Version of Record: https://www.sciencedirect.

com/science/article/pii/S0304383519306135
Manuscript_5458d102a0254482bf2e7533d705ee57

Artificial intelligence in cancer diagnosis and prognosis:

Opportunities and challenges
Shigao Huang a,#, Jie Yang b,c,#, Simon Fong b,d* and Qi Zhao a,*
a
Cancer Center, Institute of Translational Medicine ,Faculty of Health Sciences, University of Macau, Taipa, Macao
SAR, P.R. China.
b
Department of Computer and Information Science, University of Macau, Taipa, Macau, P.R. China.
c
Chongqing Industry＆Trade Polytechnic, Chongqing, China.
d
Zhuhai Institute of Advanced Technology Chinese Academy of Sciences, Zhuhai, China.
# The authors contributed equally: Shigao Huang and Jie Yang.
*Correspondence: Cancer Center, Institute of Translational Medicine,Faculty of Health Sciences,
University of Macau, Taipa, Macao. Email [email protected] or Department of Computer and
Information Science, University of Macau, Taipa, Macau, P.R. China .Email: [email protected];

Abstract: Cancer is an aggressive disease with a low median survival rate. Ironically, the treatment
process is long and very costly due to its high recurrence and mortality rates. Accurate early
diagnosis and prognosis prediction of cancer are essential to enhance the patient’s survival rate.
Developments in statistics and computer engineering over the years have encouraged many
scientists to apply computational methods such as multivariate statistical analysis to analyze the
prognosis of the disease, and the accuracy of such analyses is significantly higher than that of
empirical predictions. Furthermore, as artificial intelligence (AI), especially machine learning and
deep learning, has found popular applications in clinical cancer research in recent years, cancer
prediction performance has reached new heights. This article reviews the literature on the
application of AI to cancer diagnosis and prognosis, and summarizes its advantages. We explore
how AI assists cancer diagnosis and prognosis, specifically with regard to its unprecedented
accuracy, which is even higher than that of general statistical applications in oncology. We also
demonstrate ways in which these methods are advancing the field. Finally, opportunities and
challenges in the clinical implementation of AI are discussed. Hence, this article provides a new
perspective on how AI technology can help improve cancer diagnosis and prognosis, and continue
improving human health in the future.
Keywords: Cancer diagnosis; Prognosis prediction; Deep learning; Machine learning; Deep neural
network
Highlights:
Artificial intelligence (AI), especially machine learning and deep learning, have become quite
entrenched in clinical cancer research in recent years, and prediction performance in this field has
reached new heights.
AI is known to assist cancer diagnosis and prognosis, given its unprecedented accuracy level,
which is even higher than that of general statistical applications in oncology.

© 2019 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
An overview of how AI technology could be leveraged in this area and thereby contribute to
improved human health.
1. Introduction
In 2019, the cancer burden is estimated 1.7 million new cases and 0.6 million deaths in the
United States [1].As the incidence rate of cancer and its mortality have risen sharply, prolonging
survival and reducing local recurrence have become increasingly dependent on modern
laparoscopy surgery, robotic surgery, tumor adjuvant therapy, and other new technologies [2].
Treatment for cancer currently involves various options [3-5]. Since the 2010s, the effectiveness
of cancer treatment has improved significantly [1-5]. However, despite the profusion of new
techniques, scientifically satisfactory curative results for each stricken individual are elusive due
to uncertainties in diagnostic precision. Thus, patient-specific optimal treatment could be adopted
if an accurate prognosis could be made. Indeed, improvements in prediction accuracy could greatly
assist doctors in planning patient treatments and eliminating both the physical and mental
hardships brought on by the disease. Fundamental clinical observations can be combined with the
implementation of the traditional TNM staging approach (based on the size of the tumor (T), the
spread of the cancer into nearby lymph nodes (N), and the spread of the cancer to other body parts
(M, for metastasis)) in empirical tests, but erroneous predications of prognoses continue to pose a
bottleneck for clinicians [10]. Improvement in prognosis accuracy using state-of-art AI technology
remains a critical challenge for clinical researchers.
Technical advancements in statistics and computer software have enabled computer engineers
and health scientists to collaborate closely toward prognosis improvements using multi-factor
analysis, conventional logistic regression, and Cox analyses. The accuracy of such predictions was
found to be significantly higher than that of empirical predictions. With the implementation of AI,
scholars have lately turned to establishing models using AI algorithms to predict and diagnose
cancer. These methods currently play a major role in improving the accuracy of cancer
susceptibility, recurrence, and survival predictions.

2
 Fig. 1. Process of searching and selecting primary studies for the review conducted in this study

Cancer is difficult to diagnose at early stages or can easily relapse after treatment. Moreover,
accurate predictions of disease prognosis with high certainty are very difficult. Some cancers are
difficult to detect in their early stages due to their vague symptoms and the indistinctive tell-tale
signs on mammograms and scans. Thus, developing better predictive models using multi-variate
data and high-resolution diagnostic tools in clinical cancer research is imperative. A quick search
of the literature shows that the number of papers on cancer analysis has grown exponentially,
especially those involving AI tools and large data sets containing historical clinical cases for
training AI models [6]. Moreover, the literature reports that traditional analysis methods such as
statistical analysis and multivariate analysis are not as accurate as AI. This is particularly true when
AI is used in conjunction with sophisticated bioinformatics tools, which can significantly improve
the diagnostic, prognostic, and predictive accuracies [7-10]. A more specific concept, namely
machine learning (ML), is becoming increasingly prevalent. ML is a subset of AI, and is used to
construct predictive models that learn logical patterns from mass historical data so as to predict
the survival rate of a patient. ML has been used extensively for improving prognosis [11-13].
Prognostication is an important clinical skill, particularly for clinicians working with cancer
patients [14, 15]. ML methods have been shown to improve the accuracy of cancer susceptibility,
recurrence, and survival predictions, three aspects that are fundamental to early diagnosis and
prognosis in cancer research. ML can provide good results with regard to the clinical management
of patients [16-22]. This aspect has motivated researchers in the biomedical and bioinformatics
fields to develop more effective ML tools that can classify cancer patients into high- or low-risk
recurrence groups for refined prognosis management.
Over the years, AI has been widely used in clinical cancer research due to its feasibility and
advantages. This study selected and analyzed relevant publications from the PubMed, Google
3
Scholar, CNKI, and WANFANG databases between 1995 and 2019. In total, using matching
keywords, 3594 papers in these databases were found to be related to AI studies on cancer. Of
these, 1136 papers were found to be duplicate and excluded. leaving a total of 2458 papers. These
papers were further examined for relevance using their titles/abstracts, and 2365 papers were found
to be relevant. Using a forward citation search, we included 126 full-text papers on the use of AI
for cancer diagnosis and prognosis predictions. This process is illustrated in Fig. 1. The majority
of the papers (66) were published in the United States, followed by China (40). Most of those
papers contain keywords such as “machine learning,” “deep learning,” “artificial intelligence,” and
“prognosis.” Our comprehensive survey provides a systematic literature review on the use of AI
in cancer research, and provides important insights into the applications of AI to cancer diagnosis
and prognosis in order to spearhead cutting edge research in this field.
2. Basics of artificial intelligence
The concept of AI first emerged in 1956, the aim being to build machines that can think and
reason over complex tasks just like human beings and thereby sharing the same essential cognitive
characteristics. Since then, the field of AI has made many developments as AI theories and
implementation gradually became reality in scientific research laboratories. Figure 2 illustrates the
milestones in the field of AI research, as well as compares the advantages and disadvantages of
the different working principles of ML and deep learning (DL). These methods have different basic
characteristics and can be applied in various fields. In recent years, researchers developed these
approaches and introduced the broad learning system for application in cancer precision medicine.
We summarize these AI methods in this review and systematically show the differences among
them and their applications in cancer analysis by clinical specialists. AI research is now expanding
into various sub-branches, such as expert systems, ML, evolutionary computing, fuzzy logic,
computer vision, natural language processing, and recommendation systems. Figure 2 shows the
branches of AI application in clinical research. Fundamentally, ML uses algorithms to parse data,
learn underlying patterns, and offer insights using which decisions and predictions about real-
world events can be made. Unlike traditional hard-coded software programs that solve specific
tasks, ML uses large amounts of data to “train” and apply algorithms to dynamically learn how
certain tasks can be accomplished. DL is not an independent learning method. Two modes,
supervised and unsupervised learning methods, are available to train a deep neural network. The
exponential development of this field in recent years has given rise to unique learning methods
such as the residual network. Thus, increasingly, DL is now being regarded as a learning method
alone. However, simply put, ML is used to realize AI while DL is used to implement ML. Despite
all of its advances, the following limitations remain with regard to DL: 1) DL models require a
large amount of training data to induce an accurate model. However, in real life, certain biomedical
samples may only exist in small quantities. 2) In some fields, traditional and simple ML methods
can be used to solve problems; complex DL methods are not required.

4
Fig. 2. The concepts of AI, ML, and DL, and their relationships. This schematic outlines the milestone in AI
research, and compares the advantages and disadvantages as well as the different working principles of ML and DL.
a) ML and DL are used to implement AI, and DL is a technology used to implement ML. AI, ML, and DL were first
discovered in the 1950s, 1980s, and 2010s, respectively. b) Features of and differences between ML and DL with
regard to features engineering, execution time, interpretability, and data and hardware dependency. c) ML relies on
engineered features extracted from specific regions on the basis of expert knowledge. The workflow comprises the
following steps: data preprocessing after acquisition, features extraction, and selection and classification of algorithms.
d) DL uses localization for features engineering without region annotations. It comprises several layers in which
features extraction, selection, and ultimate classification are performed simultaneously during training. Both methods
are data-centric and involve AI, which has been applied to the field of cancer research.

3. Artificial intelligence in clinical cancer prognosis prediction

In the last few decades, various types of clinicians, ranging from specialty doctors to
paramedics, have been called upon to make clinical cancer prognosis predictions based on their
work experience. With the arrival of the digital information era, clinicians understand the
importance of using AI technology such as ML and DL as a decision support tool (Fig. 2a). Many
differences exist between ML and DL (Fig. 2b). For instance, ML (Fig. 2c) involves more
processes than DL (Fig. 2d). In our previous study, we used AI to predict brain metastasis [23]. AI
is thus being increasingly applied toward clinical cancer prognosis. Given the failure of traditional
statistical analysis to provide accurate predictions, it is difficult to predict how a patient’s cancer
will progress. Clinicians are also concerned about patients’ risk of contracting the disease, tumor
recurrence, or death after treatment. Such aspects are highly related to the choice of treatment and
curative effects. In fact, most research on clinical cancer is currently aimed at determining the
prognosis or predicting the correct outcome in response to treatment. If the prognoses of different
patients can be predicted more accurately, more precise and suitable treatments can be provided to
them; in fact, such treatments tend to be individualized or customized to patients. To date, accurate
treatment customized for a patient is very difficult to implement. However, AI can be used to

5
process and analyze multi-factor data from multiple patient examination data to predict cancer
prognosis as well as the survival time and the disease progress of patients more accurately.
While medical statistics are commonly used for cancer prognostics, the use of AI for the same
task tends to be less common. In this paper, we present a review of the literature on application of
AI to various kinds of cancer prognoses by global scientists in different study populations (Table
1). In the last decade, the number of such studies have increased rapidly in China, the United States,
and Europe. Usually, medical statistics cover methods such as area under the curve (AUC). Cancer
prognosis involves predictions of disease recurrence and patient survival, the aim being to improve
patient management[24, 25]. Enshaei A, et al. [26] compared a variety of algorithms and
classifiers with conventional logistic regression statistical approaches to demonstrate that AI may
have a role in providing prognostic and predictive data for ovarian cancer patient. Khan et al. [27]
used a number of decision tree rules, fuzzy membership functions, and inference techniques for
breast cancer survival analysis. Their performance comparisons suggested that predictions of
weighted fuzzy decision trees (wFDT) are more accurate and balanced than independently applied
crisp decision tree classifiers. Moreover, this approach showed good potential for significant
cancer prognosis prediction performance enhancement.
3.1. Breast cancer prognosis prediction
Breast cancer prognosis involves estimation of the recurrence of disease and predicting the
survival of the patient, hence resulting in improved patient management. Researchers often use
multimodal deep neural networks (DNNs) by integrating multi-dimensional data to compare of the
receiver operating characteristic (ROC) curve and AUC values. The results indicate that combining
different data types and ensemble DNN methods is an efficient way to improve human breast
cancer prognosis predictions (Fig. 3). Jhajharia et al. [28] applied principal component analysis by
preprocessing the data and extracting features in the most relevant form for training artificial neural
networks (ANNs) to learn patterns in the data for classification of new instances. Data- and
learning-based approaches can provide an effective framework for prognostic research by
accurately classifying data instances into relevant categories based on tumor severity. Ching et al.
[29] developed a new ANN framework called Cox-nnet (a neural network extension of the Cox
regression model) to predict patient prognoses from high throughput transcriptomics data. Cox-
nnet reveals much richer biological information, at both the pathway and gene levels, by analyzing
features represented in the hidden layer nodes in Cox-nnet. Bomane A, et al. [30] applied three
classifiers features selected to individually link to the cytotoxic-drug sensitivities and prognosis of
patients on breast cancer for optimizing paclitaxel-therapies in clinical practice.Sun et al. [31]
proposed a multimodal DNN by integrating multi-dimensional data (MDNNMD) for the prognosis
prediction of breast cancer. The novelty of their method lies in the design of the method’s
architecture and the fusion of multi-dimensional data. The results of the comprehensive
performance evaluation showed that the proposed method outperformed all the other prediction
methods using single-dimensional data.
Chi et al. [32] applied ANNs to survival analysis as ANNs can easily consider variable
interactions and create a non-linear prediction model thus offering more flexible predictions of
survival time than traditional methods. Their study compared the results of the ANNs for two
6
different breast cancer datasets, both of which contained nuclear morphometric features. The
results showed that ANNs can successfully predict recurrence probabilities and separate patients
with good (more than five years) and bad (less than five years) prognoses. Park et al. [33] suggested
that a semi-supervised learning model can be easily applied by medical professionals without
expending the time and effort for parameter searching in conventional models. The ease of use and
lowered search time will eventually lead to more accurate and less-invasive prognoses for breast
cancer patients. Delen et al. [34] used ANNs and decision trees along with a traditional statistical
method (logistic regression) to develop prediction models using more than 200,000 cases. They
found that the decision tree (C5) was the best predictor, showing 93.6% accuracy on the holdout
sample, followed by ANNs with 91.2% accuracy, and logistic regression models with the worst
accuracy (89.2%). Sun et al. [35] and Gevaert et al. [36] proposed a more practical strategy that
utilized both clinical and genetic marker information, which may be complementary given the
complexity of breast cancer prognosis.

7
 Table 1 AI applied to various kinds of cancer prognosis by the global scientist in different study population

Number of
Country/ Age Study
Type of Cancer Authors Year Patients in Methods Results
Region (Years) Population
Study
Prostate Cancer Kuo et al[37] 2015 Taiwan 100 75 Hospital Fuzzy Neural Network /
Zhang et al[38] 2017 USA / / TCGA SVM model Average Accuracy (66%)
Breast Cancer Sun et al [31] 2018 China 1980 61 Multimodal DNN /
Park et al [33] 2013 USA 162500 N/A SEER Semi-supervised Learning Model /
Delen et al [34] 2005 USA 433272 60.61 SEER ANN and DT Accuracy: DT (93.6%), ANN
(91.2%)
Lu et al [39] 2019 USA 82707 58.38 SEER Dynamic Gradient Boosting Machine with GA Accuracy Improved (28%)
Glioblastoma Vasudevan et al [40] 2018 India 215 N/A TCGA Neural Network Accuracy: DT (89.2%)
Bladder Cancer Tian et al [41] 2019 China 115 N/A Hospital Statistical Analysis NEDD8: Poor Prognosis Found
Hasnain et al [42] 2019 USA 3503 67.8 Hospital KNN, RF, etc Sensitivity& Specificity (>70%)
Nasopharyngeal Zhang et al [21] 2019 China 3269 45 Hospital Large Scale, Big Data Intelligence Platform EBV DNA: a Robust Biomarker for
Carcinoma NPC Prognosis
Gastric Cancer Biglarian et al [43] 2011 Iran 436 58.43±13.02 Hospital Cox Proportional Hazard, ANN TP(83.1%),
Zhu et al [44] 2013 China 289 63.20±10.75 Hospital ANN TP: ANN(85.3%)
Colorectal Cancer Bottaci et al [45] 1997 UK 334 N/A Hospital Six Neural Networks Accuracy(>80%),
mean Sensitivity(60%),
mean Specificity(88%)
Wang et al [46] 2019 China 1568 N/A SEER Semi-random Regression Tree /
Bychkov et al [47] 2018 Finland 641 N/A Hospital LSTM, Naïve Bayes, SVM Hazard Ratio(2.3); CI(95%,1.79–
3.03), AUC(0.69)
Oral Cancer Chang et al [48] 2013 Malaysia 156 N/A MOCDTBS Hybrid model of ReliefF-GA-ANFIS Accuracy(93.81%),AUC (0.9)
Lung Cancer Lynch et al [49] 2017 USA 10442 N/A SEER GBM, SVM RMSE(32,15.05) for GBM, SVM
Sepehri et al [50] 2018 France 101 N/A Hospital SVM with RFE and RF Accuracy(71%, 59%)
Yu et al [51] 2016 Italy 168 N/A Hospital Naive Bayes, SVM with Gaussian, etc /
Ovarian Cancer Lu et al [52] 2019 Taiwan 84 59.94±11.25 Both SVM HR(0.644), CI(95％,0.436-0.952)
Lu et al [53] 2019 UK 364 N/A Both Unsupervised Hierarchical Clustering RPV: A Novel Prognostic Signature
Discovered
Acharya et al [54] 2018 Singapore& 469 23-90 Hospital Fuzzy Forest Accuracy(80.60 ± 0.5%),
Malaysia Sensitivity(81.40%), Specificity
(76.30%)
Glioma Lu et al [55] 2018 Taiwan 456 N/A TCGA Improved SVM Accuracy(81.8%), ROC(0.922)
Papp et al [56] 2018 Austria 70 48±15 Hospital GA and Nelder–Mead ML methods Sensitivity (86%-98%), Specificity
(92%-95%)
Spinal Chordoma Karhade et al [57] 2018 USA 265 N/A SEER Boosted DT, SVM, ANN 5-year Survival (67.5%)
Long Bone Metastases Stein et al [58] 2015 USA 927 62±13 Hospital Multiple Additive Regression Trees /
Oral Cavity Squamous Lu et al [59] 2017 USA 115 61.0±12. Hospital RF, SVM AUC(0.72), Accuracy(70.77),
Cell Specificity(73.08),
Sensitivity(61.54)
Pancreatic Song et al [60] 2018 China 8422 59(48-69) SEER SVM, RF,DL Accuracy(81.6%±1.9%),curve(0.87)
Neuroendocrine

8
*SVM: Support Vector Machine, DNN: Deep Neural Network, ANN: Artificial Neural Network, DT: Decision Tree, GA: Genetic Algorithm Optimizer, KNN: K-Nearest Neighbor, RF: Random Forest, LSTM: Long Short-
Term Memory Network, GBM: Gradient Boosting Machines, RFE: Recursive Feature Elimination, TP: True Prediction

9
 The hybrid signature performed significantly better than other methods, including the 70-gene
signature, clinical makers alone, and the St. Gallen consensus criterion. At the 90% sensitivity
level, the hybrid signature achieved 67% specificity as compared to 47% for the 70-gene signature
and 48% for the clinical makers. The odds ratio of the hybrid signature for developing distant
metastases within five years between patients with a good prognosis signature and those with a
bad prognosis was 21.0 (95% CI: 6.5–68.3), far higher than that of either the genetic or the clinical
markers alone. Xing et al. [61] presented a general clustering-based approach called algorithm of
clustering of cancer data(ACCD) to develop a predictive system for cancer patients. Xu et al. [62]
adopted an efficient feature selection method, known as the support vector machine-based
recursive feature elimination (SVM-RFE), for gene selection and prognosis prediction. Using the
leave-one-out evaluation procedure on a gene expression dataset including 295 breast cancer
patients, they discovered a 50-gene signature, which could be combined with SVM to achieve a
superior prediction performance showing an improvement of 34, 48, and 3% in accuracy,
sensitivity, and specificity, respectively, compared with the widely used 70-gene signature. Further
analysis showed that the 50-gene signature was effective at predicting the prognoses of metastases
and distinguishing patients who should receive adjuvant therapy. Lu et al. [39] showed that their
proposed a genetic algorithm-based online gradient boosting (GAOGB) model achieved
statistically outstanding online learning effectiveness. Rohit et al. [63] used three different machine
learning methods to predict breast cancer survivability separately for each stage, and compared
them with the traditional joint models for all the stages. Wang et al. [64] showed that the synthetic
minority over-sampling technique + particle swarm optimization + C5 (SMOTE + PSO + C5)
hybrid algorithm is the best among all algorithm combinations for 5-year survivability of breast
cancer patient classification. Shukla et al. [65] developed a robust data analytical model to help
improve understanding of breast cancer survivability in presence of missing data. Beibit et al. [66]
proposed a neural network-based entity embedding approach to acquire continuous vector
representations of categorical variables to interpret categorical variables and improve prognosis
using classifiers.

10
Fig. 3. Human breast cancer prognosis prediction using multimodal DNNs by integrating multi-dimensional
data including gene expression profile, copy number alteration (CNA) profile, and clinical data. The prediction
model consists of a triple modal DNN and finally combined predictive scores from each independent model. a and b
[31] show a comparison of the ROC curve and AUC value. The results indicated that combining different data types
and ensemble DNN methods is an efficient way to improve human breast cancer prognosis prediction performance.

Medical images trained with deep learning can further improve the accuracy of cancer re-
staging. Using big data from images to establish the prognostic model, we can acquire a superior
prediction of cancer patient prognosis. Sepehri et al. [50] compared two ML pipelines to build
prognostic models exploiting clinical and 18F-FDG PET/CT radiomics features in lung cancer
patients. 18F-FDG PET/CT is a combination of positron emission tomography (PET) with 18F-
labeled fluoro-2-deoxyglucose (18F-FDG) and computed tomography (CT). They showed that
although SVM provided better accuracy than RF in the training step, RF had the highest validation
performance (71% vs. 59%). Cao et al. [67] evaluated 168 patients affected by ovarian carcinoma,
who underwent a restaging 18F-FDG PET/CT. The increased odds ratio was assessed using Cox
regression analysis testing of all lesion parameters measured by PET/CT, and the results indicated
that 18F-FDG PET/CT has important prognostic value in assessing the risk of disease progression
and mortality rate.
3.2. Gastric cancer prognosis prediction
ANN has been shown to be a more powerful statistical tool for predicting the survival rate of
gastric cancer patients compared to the Cox proportional hazard regression model. Oh et al. [68]
used a survival recurrent network (SRN) to predict survival, and the results corresponded closely
with actual survival. Thus, the SRN model was more accurate at survival prediction than the
staging defined by the American Joint Committee on Cancer (AJCC). While TNM staging is a
grouped prediction considering only tumor factors, the SRN model provides an individualized
prediction based on numerous prognostic factors; basically, patient grouping is not necessary.
11
Biglarian et al. [56], analyzed 436 registered gastric cancer patients who had had surgery between
2002 and 2007 at the Taleghani Hospital, Tehran (Iran), to predict the survival time using Cox
proportional hazard and ANN techniques. The estimated one-, two-, three-, four-, and five-year
survival rates of the patients were 77.9, 53.1, 40.8, 32.0, and 17.4%, respectively. The Cox
regression analysis revealed that the age at diagnosis, high-risk behaviors, extent of wall
penetration, distant metastasis, and tumor stage were significantly associated with the survival
rates of the patients. The true prediction of the neural network was 83.1%, and the corresponding
value for the Cox regression model was 75.0%.
Another study [57] demonstrated that the ANN model is a more powerful tool in determining
significant prognostic variables for gastric cancer patients, which are recommended for
determining the risk factors of such patients. Maroufizadeh et al. [73, 74] showed that the neural
network model is a more powerful tool in determining important variables for gastric cancer
patients compared to the conventional statistical method (Weibull regression model) [75, 76].
Alexander et al. [77] predicted disease-specific gastric cancer survival at a European institution by
using a U.S.-derived nomogram. Amiri et al. [78] assessed the application of neural networks to
survival analysis in comparison to the Kaplan–Meier and Cox proportional hazards models.
4. Artificial intelligence in cancer diagnosis
Clinicians usually rely on their personal knowledge and clinical experience when examining
patients’ signs and symptoms. This clinical information and data can be used to diagnose disease,
but the accuracy of the diagnosis cannot be guaranteed, and it is impossible to avoid mistaken
diagnoses. This aspect points to the limited ability of the human brain to integrate large amounts
of sample data. However, AI models are extremely adept at handling vast amounts of data.
Integrative processing and extraction can allow more accurate disease diagnosis due to the
efficiency and effectiveness of learning and training large samples (Fig. 4). Their practicality and
accuracy are also higher than those of expert diagnoses. DL refers to a set of computer models that
have recently been used to make unprecedented progress in the way computers extract information
from images. DL algorithms have been applied to tasks in numerous medical specialties (most
extensively, radiology and pathology), and in some cases, they have attained performance
comparable to that of human experts. Furthermore, it is possible that DL could be used to extract
information from medical images that would not be immediately apparent by human analysis alone,
and that could be used to inform on molecular status, prognosis, or treatment sensitivity [69]. A
performance comparison of prognosis and diagnosis between AI methods and pathologists is
shown in Fig. 4. The diagnostic performance of CNNs was found to be superior to that of most but
not all dermatologists.

12
Fig. 4. Performance comparison of prognosis and diagnosis between AI methods and human pathologists. a)
Bychkov et al. [47] trained a deep learning-based classifier to predict five-year disease-specific survival in a
comprehensive series of digitized tumor tissue samples of CRC stained for basic morphology. Images of H&E stained
TMA spots were obtained from 420 patients who survived and died of colorectal cancer within five years after
diagnosis. The authors compared the prognosis made by the computer and the pathologists. The long short-term
memory (LSTM) network, a DL method was assessed, and TMA spot and whole-slide level was performed by human
experts. The machine-based prognosis could extract more prognostic information from the tissue morphology of
colorectal cancer than an experienced human observer. b) Haenssle et al. [60] aimed to facilitate melanoma detection
by comparing the diagnostic performance of a CNN with that of a large group of 58 international dermatologists,
including 17 “beginners,” 11 “skilled,” and 30 “expert” doctors. The levels were classified as follows: Level-I involved
dermoscopy only, while Level-II involved dermoscopy along with clinical information and images. The diagnostic
performance of the CNN was superior to most, but not all, dermatologists. The ROC of the AUC (0.86) for the CNN
was greater than the mean ROC area of the dermatologists (0.82 and 0.79, respectively; P < 0.01). Thus, physicians
having different levels of training and experience may benefit by tapping into the image classification abilities of the
CNN.

4.1. Solid tumor diagnosis

Recently, the use of a deep convolutional neural network (DCNN) model has been shown to
improve the diagnostic accuracy of thyroid cancer by analyzing sonographic imaging data from
clinical ultrasounds [70]. The DCNN model showed similar sensitivity and improved specificity
with regard to identifying patients with thyroid cancer compared to a group of skilled radiologists.
The improved technical performance of the DCNN model warrants further investigation via
randomized clinical trials. Hu et al. [71] believed that DL models can broadly influence clinical
practice. Another study [72] developed and validated DCNN algorithms using the largest number
of images to date. Yet, the accuracy in three small-scale validation sets was not satisfactory as it
ranged from 0·857 to 0·889. Another study [73] showed that the technical performance of AI
models should be thoroughly validated in different geographic settings. Mori et al. [74] expect
giant leaps in AI applications to gastrointestinal endoscopy in the following ten years. A
convolutional neural network computer-aided detection (CNN-CAD) system based on endoscopic
images was constructed to determine invasion depth and screen patients for endoscopic resection.
The results showed high accuracy and specificity, allowing early gastric cancer to be distinguished
from deeper submucosal invasion, and minimized overestimation of invasion depth to reduce
unnecessary gastrectomies. Ichimasa K, et al [75] believed artificial intelligence significantly
13
reduced unnecessary additional surgery after endoscopic resection of T1 colorectal cancer (CRC )
without missing lymph node metastasis (LNM) positivity. A previous study [76] used a DCNN
model for classification of malignant and benign masses in digital breast tomosynthesis (DBT). A
multi-stage transfer learning approach utilizing data from similar auxiliary domains was also tested
for intermediate-stage fine-tuning [77]. Combined deep belief networks (DBNs) with extreme
learning machine (ELM) classifiers can be used to fine-tune the network weights and biases, and
when combined with a genetic algorithm (GA), they can find a suitable number of hidden layers
and neurons to promote diagnostic performance in the classification of breast cancer . Automatic
classification of perifissural nodules can make lung cancer screening more efficient and reduce the
number of follow-up visits. The results showed that the performance of this approach (AUC: 0.868)
was close to that of human experts [78]. An AdaBoosted back propagation neural network (BPNN)
using each feature type and fusing the decisions made by three classifiers to differentiate nodules
could achieve AUC values of 96.65%, 94.45%, and 81.24%, respectively, which was substantially
higher than that obtained by other approaches [79]. A novel automated pulmonary nodule detection
framework with a 2D convolutional neural network (CNN) was used to assist the CT reading
process [80]. The nodule candidate detection sensitivity was 86.42%. For the false positive
reduction, the sensitivity reached 73.4% and 74.4% at 1/8 and 1/4 FPs/scan, respectively. This
result illustrates that the proposed method could facilitate accurate pulmonary nodule detection.
Nicolas et al. [81] trained a DCNN on whole-slide images obtained from The Cancer Genome
Atlas to accurately and automatically classify them as lung adenocarcinoma (LUAD), lung
squamous cell carcinoma (LUSC), or normal lung tissue. The results showed that 6 of 10 most
commonly mutated genes in LUAD (STK11, EGFR, FAT1, SETBP1, KRAS and TP53) could be
predicted from pathology images, with AUC values ranging from 0.733 to 0.856, as measured on
a held-out population. Nam et al. [82] proposed a DL-based automatic detection algorithm, which
outperformed physicians in radiograph classification and nodule detection of malignant pulmonary
nodules on chest radiographs. The technique enhanced physicians’ performances when used as a
second reader. Other studies[83, 84] used Bayesian network meta-analysis to simultaneously re-
evaluate efficacy and safety. The results indicated certain prior distributions that yielded posterior
distributions of the parameters of interest, thus allowing clinicians and health policy makers to
make more informative decisions. Yi et al. [85] described that ML-based quantitative texture
analysis (QTA) can differentiate subclinical pheochromocytoma (sPHEO) from lipid-poor
adenoma (LPA) when adrenal incidentaloma is present. Romeo et al. [86] used the J48 classifier
as the feature selection method and obtained a diagnostic accuracy of 80% in adrenal lesions on
unenhanced MRIs, which was an improvement over that of the expert radiologist (73%). Another
study [87] used an ANN to improve the risk assessment of prostate cancer. The findings showed a
sensitivity level of 90% and an enhanced specificity by 15–28%. Thus, the results showed that
despite large amounts of input data, ANNs show promise in decreasing the number of false
positives when detecting prostate cancer. With regard to classifying skin cancer, CNNs [88] can
achieve performance on par with that of all tested experts, demonstrating the value of AI for such
tasks [89].
4.2. Non-solid tumor diagnosis

14
 The results of cluster and discriminant analyses for various types of Non-Hodgkin lymphomas
(NHLs) reveal that a combination of proliferation-associated parameters rather than a single one
facilitates better distinctions between groups of lymphomas with unequal growth characteristics in
non-solid tumors [90]. The use of DL in the automatic analysis of hematoxylin- and eosin-stained
histological images resulted in an F-measure score of 5.06% in the detection task and an
improvement of 1.09% in the accuracy measure for the classification task [91]. Moreover, a DL
algorithm called LYmph Node Assistant or LYNA could detect metastatic breast cancer in sentinel
lymph node biopsies, thus improving the pathologist’s productivity and reducing false negatives
[92]. Haenssle et al. [60] compared the diagnostic performance of a CNN with that of 58
dermatologists (30 of whom were experts). Most dermatologists performed more poorly than the
CNN (Fig. 4), thus demonstrating the advantages of AI in clinical diagnosis.

4.3. Application of artificial intelligence in cancer medical imaging

To date, AI has been utilized in many medical imaging fields such as to CT and magnetic
resonance imagery (MRI), and has facilitated accurate diagnosis and treatment. Liu et al. [93]
developed a novel DL architecture (XmasNet) based on CNNs for the classification of prostate
cancer lesions using 3D multiparametric MRI data provided by the PROSTATEx challenge. Their
proposed model outperformed 69 methods among 33 participating groups, and achieved the
second-highest AUC value (0.84) in the PROSTATEx challenge. This study showed the great
potential of DL for cancer imaging. Wang et al. [94] compared DL with a DCNN and non-DL with
scale-invariant feature transform (SIFT) image feature and bag-of-words (BoW) to distinguish
patients confirmed to have prostate cancer (PCa) from those with prostate benign conditions (BCs)
such as prostatitis or prostate benign hyperplasia (BPH). Their results suggested that DL with the
DCNN is superior to non-DL learning with SIFT image feature and BoW for differentiating fully
automated PCa patients from BCs patients. These results proved that DL can be extended to image
modalities such as MRI, CT, and PET scans of other organs.
Wang et al. [95] devised a novel DL feature and Cox proportional hazard (DL-CPH)
regression to extract effective CT-based prognostic biomarkers for high-grade serous ovarian
cancer (HGSOC). Their proposed non-invasive and preoperative model could also predict
individualized recurrence for HGSOC. Thus, the prognostic analysis method may utilize CT data
without follow-up for prognostic biomarker extraction. Medeiros et al. [96] introduced a novel DL
approach to assess fundus photographs and provide quantitative information about the amount of
neural damage, which can then be used to diagnose and stage glaucoma. In addition, the DL-based
algorithm could overcome limitations of human labeling and be applied to other areas of
ophthalmology.

5. Challenges and future outlook

AI can successfully handle complex nonlinear relationships, fault tolerance, parallel
distributed processing, and learning [97]. Given its advantages of self-adaptation, simultaneous
treatment of quantitative and qualitative knowledge, and validated results from a number of
clinical studies in multiple fields [98]. AI clearly has varied uses in the field of clinical medicine
[99]. It not only makes full use of the various aspects of clinical diversity[100, 101], but also helps
to address the current lack of objectivity and universality in expert systems[102]. The application
15
of AI can help hospitals train junior physicians in clinical diagnosis and decision-making. A
growing number of research papers are reporting about the impressive diagnostic and prognosis
performance of computer systems built using ML [103, 104]. DL techniques, in particular, are
transforming our ability to interpret imaging data [105, 106]. These results may improve sensitivity
and ensure fewer false positives than radiologists. However, they also run the risk of overfitting
the training data, resulting in brittle degraded performance in certain settings[107]. Thus, ML often
involves a tradeoff between accuracy and intelligibility. More accurate models, such as boosted
trees, random forests, and neural nets, are usually not intelligible, whereas more intelligible models,
such as logistic regression, naive-Bayes, and single decision trees, often provide significantly
worse accuracy [108]. Recent work using advanced in vivo imaging, computational

16
 Table 2 AI applied to various kinds of cancer prognosis by the global scientist in different study population

Country/ Number of
Authors Year patients in the Study population Methods Results
Region study

Sensitivity(93.4%), CI(95%,89.6-96.1)
Li et al [70] 2019 China 17627 Both DCNN
Specificity(86.1%,p<0.0001)

Esteva et al [89] 2017 USA 2032 Both Inception v3 CNN AUC (over 91%)

Sensitivity(76.47%), and
Zhu et al [76] 2019 China 203 Hospital CNN Specificity(95.56%), Overall
Accuracy(89.16%),CI(95%,90-97)

Samala et al [77] 2018 USA 2566 Both DCNN AUC(0.85±0.05)

TCGA,NCI Genomic
Coudray et al [81] 2018 USA 137 DCNN(inception v3) AUC(0.733-0.856)
Data Commons

Wu et al [109] 2018 Italy 1034 Hospital Bayesian network /

Yi et al [85] 2018 Italy 436 Hospital Decision Tree J48 Accuracy (80%)

Stephan et al [87] 2002 Germany 928 Hospital ANN Specificity Level (90%)

Lorenzo et al [90] 1999 Italy 98 Hospital Multivariate Cluster Analysis /

Convolutional
Lymphoma and IDC F-measure Score Improved (5.06%),
Nadia et al [91] 2019 Italy 374 Autoencoder,Supervised
Datasets Accuracy Improved (5.06%)
Encoder FusionNet

Tabibu et al [110] 2019 India Ensemble TCGA CNN Accuracy (92.61%)

International
International Skin
Skin Imaging Google's Inception v4 CNN Sensitivities(86.6%-88.9%, ROC
Haenssle et al [60] 2018 100 Imaging Collaboration
Collaboration architecture AUC(>0.86,P < 0.01)
(ISIC)
(ISIC)

DCNN:Deep Convolutional Neural Network, ANN: Artificial Neural Network, AUC: Area Under the Curve, Ensemble:1027 (KIRC), 303 (KIRP), and 254 (KICH) tumor slide
images.

17
modelling, and animal modelling has identified barriers in the tumor microenvironment that hinder
therapy and promote tumor progression [9]. Along with other risk factors identified from blood
counts, red cell distribution width was used in an ML-based approach to generate a clinical data-
driven prediction model that was capable of predicting acute myeloid leukemia 6–12 months
before diagnosis with high specificity (98.2%) but low sensitivity (25.7%) [111, 112]. Thus, while
the application of AI in clinical cancer is likely to increase, the following challenges should be met
in order for it to remain viable.
AI technology faces some important challenges that must be resolved to ensure its use in
cancer diagnosis and prognosis [113]. For example, medical imaging data cannot be used as input
data directly. It is crucial to extract features from the imaging data and process them. Development
and popularization of technology, in addition, the weights coefficient in the neural network models
are tested, calculated, and the confidence interval is reasonable, so medical interpretation need
further research [114]. Increasing research on ANNs will likely result in their increased use in the
field of clinical medicine. While the importance of AI to this field is recognized, the joint efforts
of computer experts and medical experts toward ensuring interdisciplinary personnel training and
collaboration are crucial. Only then can the potential of this technology be put to practical and
economic application by medical staff [115]. A more pessimistic view was offered in Ref. [116],
which referred to inherent uncertainties in medicine, and the possibility that the “black box” of
neural networks/ML applications will reduce physician skills and soon transform some sectors of
healthcare in ways that may appear to be practical and economic but with unintended negative
consequences. Another crucial issue with regard to the future of AI in medicine involves privacy
and data security assurances [117]. While recent years have witnessed much enthusiasm about the
potential of “big data” and ML-based solutions, to date, only a few examples exist to illustrate the
impact of AI on current clinical practice [10, 118-120]. Obermeyer et al. [121]showed that
attention has to shift to new statistical tools from ML to be critical for anyone practicing medicine
in the 21st century. The stimulating debate that whether AI are “smarter” than human practitioners
is largely irrelevant, and we will consistently improve our collective health by using every
information and data resource[107].
6. Conclusion
AI is slowly pervading all aspects of our lifestyle, especially medicine. The review presented
in this paper shows that researchers are rapidly acquiring a much deeper understanding of the
challenges and opportunities presented by AI as an intelligent information science in the field of
cancer diagnosis and care. The potential of AI for various types of cancer prognosis and diagnosis
is reported in this paper. But, the limit of review is that we did not include the genomics and
radiomics data applied by AI to acquire clinical precise medicine. We expect that AI-based clinical
cancer research will result in a paradigm shift in cancer treatment, thereby resulting in dramatic
improvement in patient survival due to enhanced prediction rates. Thus, it is logical to expect that
the challenges of cancer prognosis and diagnosis will be solved by advances in AI in the
foreseeable future.
CRediT authorship contribution statement

18
 Shigao Huang: Contributed to the study design, Study coordination, Writing - review &
editing. Jie Yang: Contributed to the literature search, writing, review, editing and funding
acquisition. Simon Fong: Contributed to Review & editing. Qi Zhao: Contributed to the critical
review & editing and Funding acquisition. All authors read and approved the final manuscript.
Declaration of competing interest
The authors declare no competing financial interests.
Acknowledgments
This work was funded by the Science and Technology Development Fund of Macau
(FDCT/131/2016/A3, FDCT/0015/2018/A1, FDCT/126/2014/A3) and Start-up Research Grand
(SRG2016-00082-FHS), the Multi-Year Research Grant (MYRG2019-00069-FHS, MYRG2016-
00069-FST), the intramural research program of Faculty of Health Sciences, University of Macau,
Guangzhou Science and Technology Innovation and Development of Special Funds, Grant no.
EF003/FST-FSJ/2019/GSTIC, and Grant no. EF004/FST-FSJ/2019/GSTI, key project of
Chongqing Industry&Trade Polytechnic (ZR201902,190101), and the project of science and
technology research program of Chongqing Education Commission of China(KJQN201903601).

Reference
[1] R.L. Siegel, K.D. Miller, A. Jemal, Cancer statistics, 2019, CA: A Cancer Journal for Clinicians, 69 (2019)
7-34.
[2] Simmons CPL, McMillan DC, McWilliams K, Sande TA, Fearon KC, Tuck S, Fallon MT, Laird BJ:
Prognostic Tools in Patients With Advanced Cancer: A Systematic Review. J Pain Symptom Manage 2017,
53(5):962-970 e910
[3] S. Huang, Dang, Y., Li, F., Wei, W., Ma, Y., Qiao, S., & Wang, Q, Biological intensity-modulated
radiotherapy plus neoadjuvant chemotherapy for multiple peritoneal metastases of ovarian cancer: A
case report, Oncology Letters, (2015) 1239-1243.
[4] S. Huang, Q. Zhao, Nanomedicine-combined immunotherapy for cancer, Current medicinal
chemistry, (2019).
[5] S. Huang, C.I. Fong, M. Xu, B.-n. Han, Z. Yuan, Q. Zhao, Nano-loaded natural killer cells as carriers of
indocyanine green for synergetic cancer immunotherapy and phototherapy, Journal of Innovative
Optical Health Sciences, 12 (2019) 1941002.
[6] Z. Obermeyer, E.J. Emanuel, Predicting the Future - Big Data, Machine Learning, and Clinical
Medicine, N Engl J Med, 375 (2016) 1216-1219.
[7] K.P.E. Gillies R J, Hricak H., Radiomics Images Are More than Pictures, They Are Data, Radiology, 278
(2015) 563-577.
[8] A. Allahyar, J. Ubels, J. de Ridder, A data-driven interactome of synergistic genes improves network-
based cancer outcome prediction, PLoS Comput Biol, 15 (2019) e1006657.
[9] M.J. Mitchell, R.K. Jain, R. Langer, Engineering and physical sciences in oncology: challenges and
opportunities, Nat Rev Cancer, 17 (2017) 659-675.
[10] A. Hosny, C. Parmar, J. Quackenbush, L.H. Schwartz, H. Aerts, Artificial intelligence in radiology, Nat
Rev Cancer, 18 (2018) 500-510.
[11] R.C. Deo, Machine Learning in Medicine, Circulation, 132 (2015) 1920-1930.

19
[12] S. Jha, E.J. Topol, Adapting to Artificial Intelligence: Radiologists and Pathologists as Information
Specialists, JAMA, 316 (2016) 2353-2354.
[13] D. Wong., S. Yip, Machine learning classifies cancer, nature, 555 (2018) 446–447
[14] P. Glare, C. Sinclair, M. Downing, P. Stone, M. Maltoni, A. Vigano, Predicting survival in patients with
advanced disease, Eur J Cancer, 44 (2008) 1146-1156.
[15] C.P.L. Simmons, D.C. McMillan, K. McWilliams, T.A. Sande, K.C. Fearon, S. Tuck, M.T. Fallon, B.J.
Laird, Prognostic Tools in Patients With Advanced Cancer: A Systematic Review, J Pain Symptom
Manage, 53 (2017) 962-970 e910.
[16] K. Kourou, T.P. Exarchos, K.P. Exarchos, M.V. Karamouzis, D.I. Fotiadis, Machine learning
applications in cancer prognosis and prediction, Comput Struct Biotechnol J, 13 (2015) 8-17.
[17] H.M. Zolbanin, D. Delen, A. Hassan Zadeh, Predicting overall survivability in comorbidity of cancers:
A data mining approach, Decision Support Systems, 74 (2015) 150-161.
[18] D. Chen, K. Xing, D. Henson, L. Sheng, A.M. Schwartz, X. Cheng, Developing prognostic systems of
cancer patients by ensemble clustering, J Biomed Biotechnol, 2009 (2009) 632786.
[19] C. Denkert, G. von Minckwitz, S. Darb-Esfahani, B. Lederer, B.I. Heppner, K.E. Weber, J. Budczies, J.
Huober, F. Klauschen, J. Furlanetto, W.D. Schmitt, J.-U. Blohmer, T. Karn, B.M. Pfitzner, S. Kümmel, K.
Engels, A. Schneeweiss, A. Hartmann, A. Noske, P.A. Fasching, C. Jackisch, M. van Mackelenbergh, P.
Sinn, C. Schem, C. Hanusch, M. Untch, S. Loibl, Tumour-infiltrating lymphocytes and prognosis in
different subtypes of breast cancer: a pooled analysis of 3771 patients treated with neoadjuvant
therapy, The Lancet Oncology, 19 (2018) 40-50.
[20] Y. Mintz, R. Brodie, Introduction to artificial intelligence in medicine, Minim Invasive Ther Allied
Technol, (2019) 1-9.
[21] Z. Qian, Y. Li, Y. Wang, L. Li, R. Li, K. Wang, S. Li, K. Tang, C. Zhang, X. Fan, B. Chen, W. Li,
Differentiation of glioblastoma from solitary brain metastases using radiomic machine-learning
classifiers, Cancer letters, 451 (2019) 128-135.
[22] A. Tan, H. Huang, P. Zhang, S. Li, Network-based cancer precision medicine: A new emerging
paradigm, Cancer letters, 458 (2019) 39-45.
[23] Huang, S.; Yang, J.; Fong, S.; Zhao, Q. Mining Prognosis Index of Brain Metastases Using Artificial
Intelligence. Cancers 2019, 11, 1140.
[24] Y. Dang, Li, X., Ma, Y., Li, X., Yang, T., Lu, W., & Huang, S. , 18F‑FDG‑PET/CT‑guided
intensity‑modulated radiotherapy for 42 FIGO III/IV ovarian cancer: A retrospective study, Oncology
Letters, 17 (2019) 149-158.
[25] H.S. Gao HX, Du JF, Zhang XC, Jiang N, Kang WX, Mao J, Zhao Q, Comparison of Prognostic Indices in
NSCLC Patients with Brain Metastases after Radiosurgery, Int J Biol Sci 14 (2018) 2065-2072.
[26] A. Enshaei, C.N. Robson, R.J. Edmondson, Artificial Intelligence Systems as Prognostic and Predictive
Tools in Ovarian Cancer, Annals of surgical oncology, 22 (2015) 3970-3975.
[27] S.H. Khan U, Choi J P, et al., wFDT weighted fuzzy decision trees for prognosis of breast cancer
survivability, Proceedings of the 7th Australasian Data Mining Conference, Australian Computer Society,
2008, pp. 141-152.
[28] S. Jhajharia, B.U. Department of Computer Engineering, Jaipur, India 304022, H.K.V.S.V.R. Kumar, A
neural network based breast cancer prognosis model with PCA processed features, 2016 International
Conference on Advances in Computing, Communications and Informatics (ICACCI), IEEE, Jaipur, India,
2016, pp. 1896-1901.
[29] T. Ching, X. Zhu, L.X. Garmire, Cox-nnet: An artificial neural network method for prognosis
prediction of high-throughput omics data, PLoS Comput Biol, 14 (2018) e1006076.
[30] A. Bomane, A. Gonçalves, P.J. Ballester, Paclitaxel Response Can Be Predicted With Interpretable
Multi-Variate Classifiers Exploiting DNA-Methylation and miRNA Data, Frontiers in Genetics, 10 (2019).

20
[31] D. Sun, M. Wang, A. Li, A multimodal deep neural network for human breast cancer prognosis
prediction by integrating multi-dimensional data, IEEE/ACM Trans Comput Biol Bioinform, (2018).
[32] C.C. L, S.W. N, W.W. H, Application of Artificial Neural Network-Based Survival Analysis on Two
Breast Cancer Datasets, AMIA Annual Symposium Proceedings, American Medical Informatics
Association, 2007, pp. 130.
[33] K. Park, A. Ali, D. Kim, Y. An, M. Kim, H. Shin, Robust predictive model for evaluating breast cancer
survivability, Engineering Applications of Artificial Intelligence, 26 (2013) 2194-2205.
[34] D. Delen, G. Walker, A. Kadam, Predicting breast cancer survivability: a comparison of three data
mining methods, Artif Intell Med, 34 (2005) 113-127.
[35] Y. Sun, S. Goodison, J. Li, L. Liu, W. Farmerie, Improved breast cancer prognosis through the
combination of clinical and genetic markers, Bioinformatics, 23 (2007) 30-37.
[36] O. Gevaert, F. De Smet, D. Timmerman, Y. Moreau, B. De Moor, Predicting the prognosis of breast
cancer by integrating clinical and microarray data with Bayesian networks, Bioinformatics, 22 (2006)
e184-190.
[37] R.J. Kuo, M.H. Huang, W.C. Cheng, C.C. Lin, Y.H. Wu, Application of a two-stage fuzzy neural
network to a prostate cancer prognosis system, Artif Intell Med, 63 (2015) 119-133.
[38] S. Zhang, Y. Xu, X. Hui, F. Yang, Y. Hu, J. Shao, H. Liang, Y. Wang, Improvement in prediction of
prostate cancer prognosis with somatic mutational signatures, J Cancer, 8 (2017) 3261-3267.
[39] H. Lu, H. Wang, S.W. Yoon, A dynamic gradient boosting machine using genetic optimizer for
practical breast cancer prognosis, Expert Systems with Applications, 116 (2019) 340-350.
[40] P. Vasudevan, T. Murugesan, Cancer Subtype Discovery Using Prognosis-Enhanced Neural Network
Classifier in Multigenomic Data, Technol Cancer Res Treat, 17 (2018) 1533033818790509.
[41] D.W. Tian, Z.L. Wu, L.M. Jiang, J. Gao, C.L. Wu, H.L. Hu, Neural precursor cell expressed,
developmentally downregulated 8 promotes tumor progression and predicts poor prognosis of patients
with bladder cancer, Cancer Sci, 110 (2019) 458-467.
[42] Z. Hasnain, J. Mason, K. Gill, G. Miranda, I.S. Gill, P. Kuhn, P.K. Newton, Machine learning models for
predicting post-cystectomy recurrence and survival in bladder cancer patients, PLoS One, 14 (2019)
e0210976.
[43] H.E. Biglarian A, Kazemnejad A, et al., Application of Artificial Neural Network in Predicting the
Survival Rate of Gastric Cancer Patients, Iranian journal of public health, 40 (2011).
[44] L. Zhu, W. Luo, M. Su, H. Wei, J. Wei, X. Zhang, C. Zou, Comparison between artificial neural
network and Cox regression model in predicting the survival rate of gastric cancer patients, Biomed Rep,
1 (2013) 757-760.
[45] L. Bottaci, P.J. Drew, J.E. Hartley, M.B. Hadfield, R. Farouk, P.W.R. Lee, I.M.C. Macintyre, G.S. Duthie,
J.R.T. Monson, Artificial neural networks applied to outcome prediction for colorectal cancer patients in
separate institutions, The Lancet, 350 (1997) 469-472.
[46] Y. Wang, D. Wang, X. Ye, Y. Wang, Y. Yin, Y. Jin, A tree ensemble-based two-stage model for
advanced-stage colorectal cancer survival prediction, Information Sciences, 474 (2019) 106-124.
[47] D. Bychkov, N. Linder, R. Turkki, S. Nordling, P.E. Kovanen, C. Verrill, M. Walliander, M. Lundin, C.
Haglund, J. Lundin, Deep learning based tissue analysis predicts outcome in colorectal cancer, Sci Rep, 8
(2018) 3395.
[48] K.S.A. Chang S W, Kallarakkal T G, et al., Feature Selection Methods for Optimizing Clinicopathologic
Input Variables in Oral Cancer Prognosis, Asian Pac J Cancer Prev, 12 (2011) 2659-2664.
[49] C.M. Lynch, B. Abdollahi, J.D. Fuqua, A.R. de Carlo, J.A. Bartholomai, R.N. Balgemann, V.H. van
Berkel, H.B. Frieboes, Prediction of lung cancer patient survival via supervised machine learning
classification techniques, Int J Med Inform, 108 (2017) 1-8.

21
[50] S. Sepehri, T. Upadhaya, M.-C. Desseroit, D. Visvikis, C.C. Le Rest, M.J.J.o.N.M. Hatt, Comparison of
machine learning algorithms for building prognostic models in non-small cell lung cancer using clinical
and radiomics features from 18F-FDG PET/CT images, Journal of Nuclear Medicine, 59 (2018) 328-328.
[51] K.H. Yu, C. Zhang, G.J. Berry, R.B. Altman, C. Re, D.L. Rubin, M. Snyder, Predicting non-small cell lung
cancer prognosis by fully automated microscopic pathology image features, Nat Commun, 7 (2016)
12474.
[52] T.P. Lu, K.T. Kuo, C.H. Chen, M.C. Chang, H.P. Lin, Y.H. Hu, Y.C. Chiang, W.F. Cheng, C.A. Chen,
Developing a Prognostic Gene Panel of Epithelial Ovarian Cancer Patients by a Machine Learning Model,
Cancers (Basel), 11 (2019).
[53] H. Lu, M. Arshad, A. Thornton, G. Avesani, P. Cunnea, E. Curry, F. Kanavati, J. Liang, K. Nixon, S.T.
Williams, M.A. Hassan, D.D.L. Bowtell, H. Gabra, C. Fotopoulou, A. Rockall, E.O. Aboagye, A
mathematical-descriptor of tumor-mesoscopic-structure from computed-tomography images annotates
prognostic- and molecular-phenotypes of epithelial ovarian cancer, Nat Commun, 10 (2019) 764.
[54] U.R. Acharya, A. Akter, P. Chowriappa, S. Dua, U. Raghavendra, J.E.W. Koh, J.H. Tan, S.S. Leong, A.
Vijayananthan, Y. Hagiwara, M.T. Ramli, K.H. Ng, Use of Nonlinear Features for Automated
Characterization of Suspicious Ovarian Tumors Using Ultrasound Images in Fuzzy Forest Framework,
International Journal of Fuzzy Systems, 20 (2018) 1385-1402.
[55] C.F. Lu, F.T. Hsu, K.L. Hsieh, Y.J. Kao, S.J. Cheng, J.B. Hsu, P.H. Tsai, R.J. Chen, C.C. Huang, Y. Yen, C.Y.
Chen, Machine Learning-Based Radiomics for Molecular Subtyping of Gliomas, Clin Cancer Res, 24
(2018) 4429-4436.
[56] L. Papp, N. Potsch, M. Grahovac, V. Schmidbauer, A. Woehrer, M. Preusser, M. Mitterhauser, B.
Kiesel, W. Wadsak, T. Beyer, M. Hacker, T. Traub-Weidinger, Glioma Survival Prediction with Combined
Analysis of In Vivo (11)C-MET PET Features, Ex Vivo Features, and Patient Features by Supervised
Machine Learning, J Nucl Med, 59 (2018) 892-899.
[57] A.V. Karhade, Q. Thio, P. Ogink, J. Kim, S. Lozano-Calderon, K. Raskin, J.H. Schwab, Development of
Machine Learning Algorithms for Prediction of 5-Year Spinal Chordoma Survival, World Neurosurg, 119
(2018) e842-e847.
[58] S.J. Janssen, A.S. van der Heijden, M. van Dijke, J.E. Ready, K.A. Raskin, M.L. Ferrone, F.J. Hornicek,
J.H. Schwab, 2015 Marshall Urist Young Investigator Award: Prognostication in Patients With Long Bone
Metastases: Does a Boosting Algorithm Improve Survival Estimates?, Clin Orthop Relat Res, 473 (2015)
3112-3121.
[59] C. Lu, J.S. Lewis, Jr., W.D. Dupont, W.D. Plummer, Jr., A. Janowczyk, A. Madabhushi, An oral cavity
squamous cell carcinoma quantitative histomorphometric-based image classifier of nuclear morphology
can risk stratify patients for disease-specific survival, Mod Pathol, 30 (2017) 1655-1665.
[60] H.A. Haenssle, C. Fink, R. Schneiderbauer, F. Toberer, T. Buhl, A. Blum, A. Kalloo, A.B.H. Hassen, L.
Thomas, A. Enk, L. Uhlmann, I. Reader study level, I.I.G. level, Man against machine: diagnostic
performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in
comparison to 58 dermatologists, Ann Oncol, 29 (2018) 1836-1842.
[61] K. Xing, D. Chen, D. Henson, L. Sheng, A Clustering-Based Approach to Predict Outcome in Cancer
Patients, Sixth International Conference on Machine Learning and Applications (ICMLA 2007), 2007, pp.
541-546.
[62] Z.Y. Xu X, Zou L, et al. , A gene signature for breast cancer prognosis using support vector machine,
2012 5th International Conference on BioMedical Engineering and Informatics, IEEE, 2012, pp. 928-931.
[63] R.J. Kate, R. Nadig, Stage-specific predictive models for breast cancer survivability, Int J Med Inform,
97 (2017) 304-311.
[64] K.-J. Wang, B. Makond, K.-H. Chen, K.-M. Wang, A hybrid classifier combining SMOTE with PSO to
estimate 5-year survivability of breast cancer patients, Applied Soft Computing, 20 (2014) 15-24.

22
[65] N. Shukla, M. Hagenbuchner, K.T. Win, J. Yang, Breast cancer data analysis for survivability studies
and prediction, Comput Methods Programs Biomed, 155 (2018) 199-208.
[66] B. Abdikenov, Z. Iklassov, A. Sharipov, S. Hussain, P.K. Jamwal, Analytics of Heterogeneous Breast
Cancer Data Using Neuroevolution, IEEE Access, 7 (2019) 18050-18060.
[67] F. Caobelli, P. Alongi, L. Evangelista, M. Picchio, G. Saladini, M. Rensi, O. Geatti, A. Castello, I. Laghai,
C.E. Popescu, C. Dolci, C. Crivellaro, S. Seghezzi, M. Kirienko, V. De Biasi, F. Cocciolillo, N. Quartuccio,
A.W.G. Young, Predictive value of (18)F-FDG PET/CT in restaging patients affected by ovarian carcinoma:
a multicentre study, Eur J Nucl Med Mol Imaging, 43 (2016) 404-413.
[68] S.E. Oh, M.-G. Choi, S.W. Seo, ASO Author Reflections: Use of the Survival Recurrent Network for
Prediction of Overall Survival in Patients with Gastric Cancer, Annals of surgical oncology, 25 (2018)
1153–1159.
[69] A.B. Levine, C. Schlosser, J. Grewal, R. Coope, S.J.M. Jones, S. Yip, Rise of the Machines: Advances in
Deep Learning for Cancer Diagnosis, Trends in cancer, 5 (2019) 157-169.
[70] X. Li, S. Zhang, Q. Zhang, X. Wei, Y. Pan, J. Zhao, X. Xin, C. Qin, X. Wang, J. Li, F. Yang, Y. Zhao, M.
Yang, Q. Wang, Z. Zheng, X. Zheng, X. Yang, C.T. Whitlow, M.N. Gurcan, L. Zhang, X. Wang, B.C. Pasche,
M. Gao, W. Zhang, K. Chen, Diagnosis of thyroid cancer using deep convolutional neural network models
applied to sonographic images: a retrospective, multicohort, diagnostic study, The Lancet Oncology, 20
(2019) 193-201.
[71] D. Hu, F. Peng, W. Niu, Deep convolutional neural network models for the diagnosis of thyroid
cancer, The Lancet Oncology, 20 (2019).
[72] E.J. Topol, High-performance medicine: the convergence of human and artificial intelligence, Nature
Medicine, 25 (2019) 44-56.
[73] E.J. Ha, J.H. Baek, D.G. Na, Deep convolutional neural network models for the diagnosis of thyroid
cancer, The Lancet Oncology, 20 (2019).
[74] Y. Mori, T.M. Berzin, S.E. Kudo, Artificial intelligence for early gastric cancer: early promise and the
path ahead, Gastrointest Endosc, 89 (2019) 816-817.
[75] K. Ichimasa, S.-e. Kudo, Y. Mori, M. Misawa, S. Matsudaira, Y. Kouyama, T. Baba, E. Hidaka, K.
Wakamura, T. Hayashi, T. Kudo, T. Ishigaki, Y. Yagawa, H. Nakamura, K. Takeda, A. Haji, S. Hamatani, K.
Mori, F. Ishida, H. Miyachi, Artificial intelligence may help in predicting the need for additional surgery
after endoscopic resection of T1 colorectal cancer, Endoscopy, 50 (2018) 230-240.
[76] Y. Zhu, Q.C. Wang, M.D. Xu, Z. Zhang, J. Cheng, Y.S. Zhong, Y.Q. Zhang, W.F. Chen, L.Q. Yao, P.H.
Zhou, Q.L. Li, Application of convolutional neural network in the diagnosis of the invasion depth of
gastric cancer based on conventional endoscopy, Gastrointest Endosc, 89 (2019) 806-815 e801.
[77] R.K. Samala, H.-P. Chan, L. Hadjiiski, M.A. Helvie, C.D. Richter, K.H. Cha, Breast Cancer Diagnosis in
Digital Breast Tomosynthesis: Effects of Training Sample Size on Multi-Stage Transfer Learning Using
Deep Neural Nets, IEEE Transactions on Medical Imaging, 38 (2019) 686-696.
[78] F. Ciompi, B. de Hoop, S.J. van Riel, K. Chung, E.T. Scholten, M. Oudkerk, P.A. de Jong, M. Prokop, B.
van Ginneken, Automatic classification of pulmonary peri-fissural nodules in computed tomography
using an ensemble of 2D views and a convolutional neural network out-of-the-box, Med Image Anal, 26
(2015) 195-202.
[79] Y. Xie, J. Zhang, Y. Xia, M. Fulham, Y. Zhang, Fusing texture, shape and deep model-learned
information at decision level for automated classification of lung nodules on chest CT, Information
Fusion, 42 (2018) 102-110.
[80] H. Xie, D. Yang, N. Sun, Z. Chen, Y. Zhang, Automated pulmonary nodule detection in CT images
using deep convolutional neural networks, Pattern Recognition, 85 (2019) 109-119.
[81] N. Coudray, P.S. Ocampo, T. Sakellaropoulos, N. Narula, M. Snuderl, D. Fenyo, A.L. Moreira, N.
Razavian, A. Tsirigos, Classification and mutation prediction from non-small cell lung cancer
histopathology images using deep learning, Nat Med, 24 (2018) 1559-1567.
23
[82] J.G. Nam, S. Park, E.J. Hwang, J.H. Lee, K.-N. Jin, K.Y. Lim, T.H. Vu, J.H. Sohn, S. Hwang, J.M. Goo,
Development and validation of deep learning–based automatic detection algorithm for malignant
pulmonary nodules on chest radiographs, Radiology, 290 (2018) 218-228.
[83] A.G. F. Passiglia, S. Rizzo, L. Incorvaia, A. Listi, V. Bazan, A. Russo, Looking for the best immune-
checkpoint inhibitor in pre-treated NSCLC patients: An indirect comparison between nivolumab,
pembrolizumab and atezolizumab, Int J Cancer, 142 (2018) 1277-1284.
[84] L.L. Y. Wu, Y. Shen, H. Wu, Comparison between PD-1/PD-L1 inhibitors (nivolumab, pembrolizumab,
and atezolizumab) in pretreated NSCLC patients: Evidence from a Bayesian network model, Int J Cancer,
143 (2018) 3038-3040.
[85] X. Yi, X. Guan, C. Chen, Y. Zhang, Z. Zhang, M. Li, P. Liu, A. Yu, X. Long, L. Liu, B.T. Chen, C. Zee,
Adrenal incidentaloma: machine learning-based quantitative texture analysis of unenhanced CT can
effectively differentiate sPHEO from lipid-poor adrenal adenoma, J Cancer, 9 (2018) 3577-3582.
[86] V. Romeo, S. Maurea, R. Cuocolo, M. Petretta, P.P. Mainenti, F. Verde, M. Coppola, S. Dell'Aversana,
A. Brunetti, Characterization of Adrenal Lesions on Unenhanced MRI Using Texture Analysis: A Machine-
Learning Approach, J Magn Reson Imaging, 48 (2018) 198-204.
[87] C. Stephan, K. Jung, H. Cammann, B. Vogel, B. Brux, G. Kristiansen, B. Rudolph, S. Hauptmann, M.
Lein, D. Schnorr, P. Sinha, S.A. Loening, An artificial neural network considerably improves the diagnostic
power of percent free prostate-specific antigen in prostate cancer diagnosis: results of a 5-year
investigation, Int J Cancer, 99 (2002) 466-473.
[88] Y. LeCun, Y. Bengio, G. Hinton, Deep learning, Nature, 521 (2015) 436-444.
[89] A. Esteva, B. Kuprel, R.A. Novoa, J. Ko, S.M. Swetter, H.M. Blau, S. Thrun, Dermatologist-level
classification of skin cancer with deep neural networks, Nature, 542 (2017) 115-118.
[90] C.A. Leoncini L, Megha T, et al., Expression of p34cdc2 and cyclins A and B compared to other
proliferative features of non‐Hodgkin's lymphomas A multivariate cluster analysis, International
journal of cancer, 83 (1999) 203-209.
[91] N. Brancati, G. De Pietro, M. Frucci, D. Riccio, A Deep Learning Approach for Breast Invasive Ductal
Carcinoma Detection and Lymphoma Multi-Classification in Histological Images, IEEE Access, 7 (2019)
44709-44720.
[92] Y. Liu, T. Kohlberger, M. Norouzi, G.E. Dahl, J.L. Smith, A. Mohtashamian, N. Olson, L.H. Peng, J.D.
Hipp, M.C. Stumpe, Artificial Intelligence-Based Breast Cancer Nodal Metastasis Detection, Arch Pathol
Lab Med, (2018).
[93] Y. Xu, A. Hosny, R. Zeleznik, C. Parmar, T. Coroller, I. Franco, R.H. Mak, H. Aerts, Deep Learning
Predicts Lung Cancer Treatment Response from Serial Medical Imaging, Clin Cancer Res, 25 (2019) 3266-
3275.
[94] X. Wang, W. Yang, J. Weinreb, J. Han, Q. Li, X. Kong, Y. Yan, Z. Ke, B. Luo, T. Liu, L. Wang, Searching
for prostate cancer by fully automated magnetic resonance imaging classification: deep learning versus
non-deep learning, Sci Rep, 7 (2017) 15415.
[95] S. Wang, Z. Liu, Y. Rong, B. Zhou, Y. Bai, W. Wei, W. Wei, M. Wang, Y. Guo, J. Tian, Deep learning
provides a new computed tomography-based prognostic biomarker for recurrence prediction in high-
grade serous ovarian cancer, Radiother Oncol, 132 (2019) 171-177.
[96] F. A.Medeiros, A. A.Jammal, A. C.Thompson, From Machine to Machine: An OCT-Trained Deep
Learning Algorithm for Objective Quantification of Glaucomatous Damage in Fundus Photographs,
Ophthalmology, 4 (2019) 513-521.
[97] He, J., et al. The practical implementation of artificial intelligence technologies in medicine. Nat
Med 25, 30-36 (2019).
[98] S. Jin, B. Wang, Y. Zhu, W. Dai, P. Xu, C. Yang, Y. Shen, D. Ye, Log Odds Could Better Predict Survival
in Muscle-Invasive Bladder Cancer Patients Compared with pN and Lymph Node Ratio, J Cancer, 10
(2019) 249-256.
24
[99] R. You, Y.P. Liu, M. Lin, P.Y. Huang, L.Q. Tang, Y.N. Zhang, Y. Pan, W.L. Liu, W.B. Guo, X. Zou, K.M.
Zhao, T. Kang, L.Z. Liu, A.H. Lin, M.H. Hong, H.Q. Mai, M.S. Zeng, M.Y. Chen, Relationship of circulating
tumor cells and Epstein-Barr virus DNA to progression-free survival and overall survival in metastatic
nasopharyngeal carcinoma patients, International journal of cancer, 145 (2019) 2873-2883.
[100] E.C. G.M. Haag, A. Ahadova, T. Schmidt, L. Sisic, S. Blank, U. Heger, L. Apostolidis, A.K. Berger, C.
Springfeld, F. Lasitschka, D. Jager, M. von Knebel Doeberitz, M. Kloor, Prognostic significance of
microsatellite-instability in gastric and gastroesophageal junction cancer patients undergoing
neoadjuvant chemotherapy, Int J Cancer, 144 (2019) 1697-1703.
[101] R.O. Alabi, M. Elmusrati, I. Sawazaki-Calone, L.P. Kowalski, C. Haglund, R.D. Coletta, A.A. Mäkitie, T.
Salo, I. Leivo, A. Almangush, Machine learning application for prediction of locoregional recurrences in
early oral tongue cancer: a Web-based prognostic tool, Virchows Archiv, 475 (2019) 489-497.
[102] B.L.H. M.H. Pedersen, S. Ehmsen, H.C. Beck, T.P. Conrads, M. Bak, H.J. Ditzel, R. Leth-Larsen,
CYPOR is a novel and independent prognostic biomarker of recurrence-free survival in triple-negative
breast cancer patients, Int J Cancer, 144 (2019) 631-640.
[103] P. Ferroni, F.M. Zanzotto, S. Riondino, N. Scarpato, F. Guadagni, M. Roselli, Breast Cancer
Prognosis Using a Machine Learning Approach, Cancers, 11 (2019) 328.
[104] N. Jiang, X. Xu, Exploring the survival prognosis of lung adenocarcinoma based on the cancer
genome atlas database using artificial neural network, Medicine, 98 (2019) e15642.
[105] G. Chartrand, P.M. Cheng, E. Vorontsov, M. Drozdzal, S. Turcotte, C.J. Pal, S. Kadoury, A. Tang,
Deep Learning: A Primer for Radiologists, Radiographics, 37 (2017) 2113-2131.
[106] K. Lan, D.T. Wang, S. Fong, L.S. Liu, K.K.L. Wong, N. Dey, A Survey of Data Mining and Deep
Learning in Bioinformatics, J Med Syst, 42 (2018) 139.
[107] J.H. Chen, S.M. Asch, Machine learning and prediction in medicine—beyond the peak of inflated
expectations, The New England journal of medicine, 376 (2017) 2507.
[108] R. Caruana, Y. Lou, J. Gehrke, P. Koch, M. Sturm, N. Elhadad, Intelligible Models for HealthCare::
Predicting Pneumonia Risk and Hospital 30-day Readmission, Proceedings of the 21th ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining - KDD '15, 2015, pp. 1721-1730.
[109] Y. Wu, L. Lin, Y. Shen, H. Wu, Comparison between PD-1/PD-L1 inhibitors (nivolumab,
pembrolizumab, and atezolizumab) in pretreated NSCLC patients: Evidence from a Bayesian network
model, Int J Cancer, 143 (2018) 3038-3040.
[110] S. Tabibu, P.K. Vinod, C.V. Jawahar, A deep learning approach for Pan-Renal Cell Carcinoma
classification and survival prediction from histopathology images, bioRxiv, (2019) 559401.
[111] S. Abelson, G. Collord, S.W.K. Ng, O. Weissbrod, N. Mendelson Cohen, E. Niemeyer, N. Barda, P.C.
Zuzarte, L. Heisler, Y. Sundaravadanam, R. Luben, S. Hayat, T.T. Wang, Z. Zhao, I. Cirlan, T.J. Pugh, D.
Soave, K. Ng, C. Latimer, C. Hardy, K. Raine, D. Jones, D. Hoult, A. Britten, J.D. McPherson, M. Johansson,
F. Mbabaali, J. Eagles, J.K. Miller, D. Pasternack, L. Timms, P. Krzyzanowski, P. Awadalla, R. Costa, E.
Segal, S.V. Bratman, P. Beer, S. Behjati, I. Martincorena, J.C.Y. Wang, K.M. Bowles, J.R. Quiros, A.
Karakatsani, C. La Vecchia, A. Trichopoulou, E. Salamanca-Fernandez, J.M. Huerta, A. Barricarte, R.C.
Travis, R. Tumino, G. Masala, H. Boeing, S. Panico, R. Kaaks, A. Kramer, S. Sieri, E. Riboli, P. Vineis, M.
Foll, J. McKay, S. Polidoro, N. Sala, K.T. Khaw, R. Vermeulen, P.J. Campbell, E. Papaemmanuil, M.D.
Minden, A. Tanay, R.D. Balicer, N.J. Wareham, M. Gerstung, J.E. Dick, P. Brennan, G.S. Vassiliou, L.I.
Shlush, Prediction of acute myeloid leukaemia risk in healthy individuals, Nature, 559 (2018) 400-404.
[112] A. Dart, How to predict the future, Nature Review Cancer, 18 (2018) 529.
[113] D.R. Thurtle, D.C. Greenberg, L.S. Lee, H.H. Huang, P.D. Pharoah, V.J. Gnanapragasam, Individual
prognosis at diagnosis in nonmetastatic prostate cancer: Development and external validation of the
PREDICT Prostate multivariable model, PLoS Med, 16 (2019) e1002758.
[114] H. Feng, Z.Y. Gu, Q. Li, Q.H. Liu, X.Y. Yang, J.J. Zhang, Identification of significant genes with poor
prognosis in ovarian cancer via bioinformatical analysis, J Ovarian Res, 12 (2019) 35.
25
[115] D. Qian, H. Liu, X. Wang, J. Ge, S. Luo, E.F. Patz, Jr., P.G. Moorman, L. Su, S. Shen, D.C. Christiani, Q.
Wei, Potentially functional genetic variants in the complement-related immunity gene-set are associated
with non-small cell lung cancer survival, Int J Cancer, 144 (2019) 1867-1876.
[116] F. Cabitza, R. Rasoini, G.F. Gensini, Unintended Consequences of Machine Learning in Medicine,
JAMA, 318 (2017) 517-518.
[117] E.J. Topol, High-performance medicine: the convergence of human and artificial intelligence, Nat
Med, 25 (2019) 44-56.
[118] H. Frohlich, R. Balling, N. Beerenwinkel, O. Kohlbacher, S. Kumar, T. Lengauer, M.H. Maathuis, Y.
Moreau, S.A. Murphy, T.M. Przytycka, M. Rebhan, H. Rost, A. Schuppert, M. Schwab, R. Spang, D.
Stekhoven, J. Sun, A. Weber, D. Ziemek, B. Zupan, From hype to reality: data science enabling
personalized medicine, BMC Med, 16 (2018) 150.
[119] G.S. Handelman, H.K. Kok, R.V. Chandra, A.H. Razavi, M.J. Lee, H. Asadi, eDoctor: machine learning
and the future of medicine, J Intern Med, 284 (2018) 603-619.
[120] D.N.L. Yeung S, Fei-Fei L, et al., Bedside computer vision–moving artificial intelligence from driver
assistance to patient safety, N Engl J Med, 378 (2018) 1269-1271.
[121] Z. Obermeyer, E.J. Emanuel, Predicting the Future - Big Data, Machine Learning, and Clinical
Medicine, The New England journal of medicine, 375 (2016) 1216-1219.

26