Papers by Ana Maria Mendonça
Scientific Reports, Apr 21, 2022
The coronavirus disease 2019 (COVID-19) pandemic has impacted healthcare systems across the world... more The coronavirus disease 2019 (COVID-19) pandemic has impacted healthcare systems across the world. Chest radiography (CXR) can be used as a complementary method for diagnosing/following COVID-19 patients. However, experience level and workload of technicians and radiologists may affect the decision process. Recent studies suggest that deep learning can be used to assess CXRs, providing an important second opinion for radiologists and technicians in the decision process, and superhuman performance in detection of COVID-19 has been reported in multiple studies. In this study, the clinical applicability of deep learning systems for COVID-19 screening was assessed by testing the performance of deep learning systems for the detection of COVID-19. Specifically, four datasets were used: (1) a collection of multiple public datasets (284.793 CXRs); (2) BIMCV dataset (16.631 CXRs); (3) COVIDGR (852 CXRs) and 4) a private dataset (6.361 CXRs). All datasets were collected retrospectively and consist of only frontal CXR views. A ResNet-18 was trained on each of the datasets for the detection of COVID-19. It is shown that a high dataset bias was present, leading to high performance in intradataset train-test scenarios (area under the curve 0.55-0.84 on the collection of public datasets). Significantly lower performances were obtained in interdataset train-test scenarios however (area under the curve > 0.98). A subset of the data was then assessed by radiologists for comparison to the automatic systems. Finetuning with radiologist annotations significantly increased performance across datasets (area under the curve 0.61-0.88) and improved the attention on clinical findings in positive COVID-19 CXRs. Nevertheless, tests on CXRs from different hospital services indicate that the screening performance of CXR and automatic systems is limited (area under the curve < 0.6 on emergency service CXRs). However, COVID-19 manifestations can be accurately detected when present, motivating the use of these tools for evaluating disease progression on mild to severe COVID-19 patients. First identified in late 2019, the coronavirus disease 2019 (COVID-19) is caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). While the majority of cases causes only mild symptoms, COVID-19 can cause difficulty breathing, pneumonia, acute respiratory distress syndrome (ARDS) and ultimately death. The SARS-CoV-2 can be easily transmitted, which makes the identification of infected individuals of the utmost importance in the containment of the pandemic 1. Reverse transcription polymerase chain reaction (RT-PCR) is the reference standard method in COVID-19 diagnosis but can present significant turnaround time and remains subject to potential shortage. Lateral flow tests (LFT), which allow much faster turnaround time, suffer however from limited and highly variable sensitivities (38.32-99.19%) 2. Furthermore, the development of new variants of SARS-CoV-2 can have an impact on the sensitivity of both RT-PCR and LFTs 3. Given the involvement of the respiratory airways in COVID-19, chest radiography (CXR) was initially proposed as an alternative screening method. While there is no single radiological feature that is indicative of COVID-19, changes include ground glass, coarse horizontal linear opacities and consolidation, most often with bilateral involvement 4. However, most COVID-19 patients do not develop pneumonia and present a normal CXR, which significantly lowers the screening value of CXR 5. Nevertheless, it has been proposed that people with severe respiratory symptoms could be quickly screened with CXR to distinguish between COVID-19 and
IEEE Access, 2020
Proliferative diabetic retinopathy (PDR) is an advanced diabetic retinopathy stage, characterized... more Proliferative diabetic retinopathy (PDR) is an advanced diabetic retinopathy stage, characterized by neovascularization, which leads to ocular complications and severe vision loss. However, the available DR-labeled retinal image datasets have a small representation of images of the severest DR grades, and thus there is lack of PDR cases for training DR grading models. Additionally, the criteria for labelling these images in the publicly available datasets is not always clear, with some images which do not show typical PDR lesions being labeled as PDR due to the presence of photo-coagulation treatment and laser marks. This problem, together with the datasets' high class imbalance, leads to a limited variability of the samples, which the typical data augmentation and class balancing cannot fully mitigate. We propose a heuristic-based data augmentation scheme based on the synthesis of neovessel (NV)-like structures that compensates for the lack of PDR cases in DR-labeled datasets. The proposed neovessel generation algorithm relies on the general knowledge of common location and shape of these structures. NVs are generated and introduced in pre-existent retinal images which can then be used for enlarging deep neural networks' training sets. The data augmentation scheme was tested on multiple datasets, and allows to improve the model's capacity to detect NVs.
Investigative Ophthalmology & Visual Science, 2018
Purpose : Although the choroidal thickness (ChT) is a pertinent sign to assess ocular health, its... more Purpose : Although the choroidal thickness (ChT) is a pertinent sign to assess ocular health, its analysis is not common because it takes time to manually segment it. Recent imaging developments in Optical Coherence Tomography (OCT) enable a better observation of deeper structures of the eye, like the choroid. We developed an application that automatically estimates ChT, allowing the ophthalmologist to promptly analyze the data and reducing the subjectivity and time consumption of the manual segmentation. Methods : To estimate the ChT, we used Enhanced Depth Imaging OCT B-scans from a Spectralis system (Heidelberg Enginnering, Germany). The automatic methodology can be split into three main parts: preprocessing, delineation of both choroidal limits-the Bruch's Membrane (BM) and Choroidal-Scleral Interface (CSI)-and finally ChT calculation. The preprocessing includes: contrast adjustment (different for each limit), reduction of the shadows cast by retinal vessels and the reduction of the speckle noise using a Stationary Wavelet Transform. The BM and CSI are delineated resorting to a minimum weight path algorithm (Fig. 1). The ChT is the distance between the two delineated limits in the scanned area. Measurements in a series of B-scans can be interpolated and mapped in the fundus image, as shown in Fig. 2; an application was developed for this purpose. A manual segmentation of BM and CSI was outlined by 2 OCT expert ophthalmologists in 8 macular volume videos with 19 B-scans each. Results : The mean absolute errors (MAE) between the automatic and each manual segmentation (7,5±3,2% and 7,9±4,4%) were comparable to the differences between the two manual segmentations (7,8±3,6%).To assess precision, differences were calculated between interpolations of the ChT of orthogonal series of B-scans. The difference of the automatic ChT (4,4±4,5%) is lower than the differences in the manual segmentations (5,5±4,5% and 6,1±5,0%). Conclusions : The automatic estimation of the ChT was successful, with a higher precision than the manual. The developed application allows the physician to easily access the ChT profile information, as well as, if necessary, perform a manual correction to the automatic segmentation.
Medical Image Analysis, 2019
challenge. This paper outlines the challenge, its organization, the dataset used, evaluation meth... more challenge. This paper outlines the challenge, its organization, the dataset used, evaluation methods and results of top performing participating solutions. We observe that the top performing approaches utilize a blend of clinical information, data augmentation, and the ensemble of models. These findings have the potential to enable new developments in retinal image analysis and image-based DR screening in particular.
Central serous chorioretinopathy is a retinal disease in which there is a leakage of fluid into t... more Central serous chorioretinopathy is a retinal disease in which there is a leakage of fluid into the subretinal space through a retinal pigment epithelium lesion that may cause a serous detachment of the neurosensory retina. Fluorescein angiography images allow the identification of these leaks. In this type of images, the lesions and the blood vessels appear bright and the remaining anatomical structures of the retina appear dark. The area related to the leakage increases throughout the angiographic sequences and, in general, the leakage can only be visualized completely in the later phases of the exam. In this work, computational methods of image processing and image analysis are used for the detection, characterization, and determination of the size progression of dye leaks along the angiographic sequences. These methods were integrated into a computer-aided diagnosis tool. To the best of our knowledge, a computer-aided diagnosis tool that allows the automatic characterization of leakage of central serous chorioretinopathy in fluorescein angiography images is not described in the literature. The central serous chorioretinopathy leakage segmentation problem is similar to leakage diseases like the diabetic macular edema, the macular retinopathy or the choroidal neovascularization. The segmentation methods for these three diseases are divided into three main categories: comparative by subtraction of images, comparative with classification and saliency detection. The main challenges to characterize the leakage are the difference in luminosity between images of the angiography sequence, the similar pixel intensities of the leaks and the vessels, and the late staining of the optic disc. The comparative methods by subtraction of images are used for the automatic characterization of the leakage of central serous chorioretinopathy in angiographic sequences because they use temporal information. As the leakage area grows during examination, temporal information helps with identification. Furthermore, algorithm steps were introduced to reduce the influence of anatomical elements (such as the background, vessels, and optic disc) in localization and segmentation of leaks. The steps are frame selection, image denoising, image registration, vessel segmentation, candidate selection in early frames, vessel inpainting, IV optic disc detection, intensity normalization, background removal, image subtraction and leakage segmentation. The leakage segmentation is processed in three phases. First, candidates are selected in the subtraction image with the Otsu method. Second, the region growing algorithm is applied to the candidate regions to segment leaks in the last frame of the sequence. Finally, the segmentation of the leaks in the remaining frames of the sequence is achieved through an algorithm of active contours, i.e., in each frame, the leaks are segmented using as input the contours of the segmented leaks in the frame immediately later. As all the leaks are segmented in all frames of the sequence, the size progression of the leak over time is also analyzed. The results were compared with the available manual annotations. They are adequate for the training set with 100 % detection and mean Jaccard index of 0.72 ± 0.17. In the test set, the results are also satisfactory. The manual annotations of the test set corresponded to the most deviant cases and even so a mean Jaccard index of 0.69 ± 0.12 was achieved. The detection only failed for the smaller leakages and when there are leakages with very different intensity means in the same image. The segmentation problems happen when the leaks have prolongated angiographic haze regions.
The Visual Computer, 2020
IJCNN'01. International Joint Conference on Neural Networks. Proceedings (Cat. No.01CH37222)
Page 1. A Neural Network Approach for the Automatic Detection of Microaneurysms in Retinal Angiog... more Page 1. A Neural Network Approach for the Automatic Detection of Microaneurysms in Retinal Angiograms Mohamed Kamel&amp;amp;amp;amp;amp;quot;), Saeid Belkassim&amp;amp;amp;amp;amp;#x27;&amp;amp;amp;amp;amp;quot; &amp;amp;amp;amp;amp;#x27;&amp;amp;amp;amp;amp;quot;Dept. of Systems Design Engineering, University of Waterloo, Waterloo, mkamel, belkassim { @watfast.uwaterloo.ca} ...
This paper presents an automatic application that provides several retinal image analysis functio... more This paper presents an automatic application that provides several retinal image analysis functionalities, namely vessel segment atio , vessel width estimation, artery/vein classification and optic dis c segmentation. A pipeline of these methods allows the computation of importa nt vessel related indexes, namely the Central Retinal Arteriolar Equiv alent (CRAE), Central Retinal Venular Equivalent (CRVE) and Arteriolarto-Venular Ratio (AVR), as well as various geometrical features associat ed with vessel bifurcations.
The Arteriolar-to-Venular Ratio (AVR) is commonly used in s tudies for the diagnosis of diseases ... more The Arteriolar-to-Venular Ratio (AVR) is commonly used in s tudies for the diagnosis of diseases such as diabetes, hypertension or cardio-vascular pathologies. This paper presents an automatic approach for t e estimation of the Arteriolar-to-Venular Ratio (AVR) in retinal im ages. The proposed method includes vessel segmentation, vessel caliber estimation, optic disc detection, region of interest determination, arte ry/vein classification and AVR calculation. The method was assessed using the i mages of the INSPIRE-AVR database. A mean error of 0.05 was obtained w hen the method’s results were compared with reference AVR values pr ovided with this dataset, thus demonstrating the adequacy of the propos ed solution for AVR estimation.
The work presented in this thesis was accomplished between 2010 and 2014. This period of time was... more The work presented in this thesis was accomplished between 2010 and 2014. This period of time was thoroughly different form the rest of my life both personally and professionally. I am writing these lines to express my gratitude to those whose support was undeniably holding my back to resist the hard challenging path throughout the last four years. First of all, I would like to thank Professor Ana Maria Mendonça and Professor Aurélio Campilho, my supervisors, who helped me to start my career form the very beginning and provided me a great scientific supervision during the whole period. Their interesting ideas always encouraged me to keep looking for solutions and whenever I felt down their efforts to keep me motivated were truly effective. Their deep knowledge in biomedical image analysis has been shedding light on the dark parts of the path. I would also like to thank Dr. Susana Penas form Centro Hospitalar São João and Professor Jorge Polónia from Faculdade de Medicina, Universidade do Porto for making the clinical data available for this work. The financial support from the FCT-Fundação para a Ciência e a Tecnologia, Portugal with the grant Ref. SFRH /BD/73376/2010 is also greatly acknowledged. I was lucky to enter to a very strong research group at Instituto de Engenharia Biomédica (INEB). I had very constructive discussions with my colleagues and they were kindly helpful to me to resolve the issues I confronted during the accomplishment of my work. During my stay at Porto, my friends' company helped me not to be emotionally affected by being so far from homeland. I would like to thank them all for their friendship, especially Ali, Mohammad and Mohsen who have been there for me through thick and thin. Last, but not least, I express my deepest thanks to my parents and my brother for their unconditional support through the last years in so many ways. I cannot thank enough my wife for her love and support, without which I would never have completed this work. vii viii "The soul, fortunately, has an interpreter-often an unconscious, but still a truthful interpreter-in the eye."
In this paper, we explore two different retinal vessel segmentationmethods for the reliable estim... more In this paper, we explore two different retinal vessel segmentationmethods for the reliable estimation of vessels caliber in retinal images inorder to assess vascular changes as an aid for the diagnosis of the ocularmanifestations of several systemic diseases, namely diabetic retinopathyand hypertensive retinopathy
Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, 2018
We propose UOLO, a novel framework for the simultaneous detection and segmentation of structures ... more We propose UOLO, a novel framework for the simultaneous detection and segmentation of structures of interest in medical images. UOLO consists of an object segmentation module which intermediate abstract representations are processed and used as input for object detection. The resulting system is optimized simultaneously for detecting a class of objects and segmenting an optionally different class of structures. UOLO is trained on a set of bounding boxes enclosing the objects to detect, as well as pixel-wise segmentation information, when available. A new loss function is devised, taking into account whether a reference segmentation is accessible for each training image, in order to suitably backpropagate the error. We validate UOLO on the task of simultaneous optic disc (OD) detection, fovea detection, and OD segmentation from retinal images, achieving state-of-the-art performance on public datasets.
VipIMAGE 2017, 2017
The automatic assessment of visual quality on images of the eye fundus is an important task in re... more The automatic assessment of visual quality on images of the eye fundus is an important task in retinal image analysis. A novel quality assessment technique is proposed in this paper. We propose to compute Mean-Subtracted Contrast-Normalized (MSCN) coefficients on local spatial neighborhoods of a given image and analyze their distribution. It is known that for natural images, such distribution behaves normally, while distortions of different kinds perturb this regularity. The combination of MSCN coefficients with a simple measure of local contrast allows us to design a simple but effective retinal image quality assessment algorithm that successfully discriminates between good and low-quality images, while delivering a meaningful quality score. The proposed technique is validated on a recent database of quality-labeled retinal images, obtaining results aligned with state-of-the-art approaches at a low computational cost.
2019 16th International Conference on Machine Vision Applications (MVA), 2019
Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness in the developed ... more Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness in the developed world. With the increasing number of diabetic patients there is a growing need of an automated system for DR detection. We propose EyeWeS, a general methodology that enables the conversion of any pre-trained convolutional neural network into a weakly-supervised model while at the same time achieving an increased performance and efficiency. Via EyeWeS, we are able to design a new family of methods that can not only automatically detect DR in eye fundus images, but also pinpoint the regions of the image that contain lesions, while being trained exclusively with image labels. EyeWeS improved the results of Inception V3 from 94.9% Area Under the Receiver Operating Curve (AUC) to 95.8% AUC while maintaining only approximately 5% of the Inception V3's number of parameters. The same model is able to achieve 97.1% AUC in a cross-dataset experiment. In the same cross-dataset experiment we also show that EyeWeS Inception V3 is effectively detecting microaneurysms and small hemorrhages as the indication of DR.
IEEE transactions on medical imaging, Jan 2, 2017
In medical image analysis applications, the availability of large amounts of annotated data is be... more In medical image analysis applications, the availability of large amounts of annotated data is becoming increasingly critical. However, annotated medical data is often scarce and costly to obtain. In this paper, we address the problem of synthesizing retinal color images by applying recent techniques based on adversarial learning. In this setting, a generative model is trained to maximize a loss function provided by a second model attempting to classify its output into real or synthetic. In particular, we propose to implement an adversarial autoencoder for the task of retinal vessel network synthesis. We use the generated vessel trees as an intermediate stage for the generation of color retinal images, which is accomplished with a Generative Adversarial Network. Both models require the optimization of almost everywhere differentiable loss functions, which allows us to train them jointly. The resulting model offers an end-to-end retinal image synthesis system capable of generating as...
In vivo observation and tracking of cell division in the Arabidopsis thaliana root meristem, by t... more In vivo observation and tracking of cell division in the Arabidopsis thaliana root meristem, by time-lapse confocal microscopy, is central to biology research. This paper discusses an automatic cell segmentation method, which selects the best cell candidates from a starting watershed segmentation. The selection of individual cells is obtained using a Support Vector Machine (SVM) classifier, based on the shape and edge strength of the cells' contour. The result is an improved segmentation, which is largely pruned of badly segmented cells.
2008 5th IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Proceedings, ISBI, 2008
In vivo observation and tracking of cell division in the Arabidopsis thaliana root meristem, by t... more In vivo observation and tracking of cell division in the Arabidopsis thaliana root meristem, by time-lapse confocal microscopy, is central to biology research. The research herein described is based on large amount of image data, which must be analyzed to determine the location and state of cells. The possibility of automating the process of cell detection/marking is an important step to provide research tools to the biologists in order to ease the search for a special event as cell division. This paper discusses an automatic cell segmentation method, which selects the best cell candidates from a starting watershed based image segmentation. The selection of individual cells is obtained using a Support Vector Machine (SVM) classifier, based on the shape and edge strength of the cells' contour. The resulting segmentation is largely pruned of badly segmented cells, which can reduce the false positive detection of cell division. This is a good result on its own and a starting point for improvement of cell segmentation methodology.
Medical Image Computing and Computer Assisted Intervention – MICCAI 2018, 2018
This paper introduces a novel strategy for the task of simultaneously locating two key anatomical... more This paper introduces a novel strategy for the task of simultaneously locating two key anatomical landmarks in retinal images of the eye fundus, namely the optic disc and the fovea. For that, instead of attempting to classify each pixel as belonging to the background, the optic disc, or the fovea center, which would lead to a highly class-imbalanced setting, the problem is reformulated as a pixelwise regression task. The regressed quantity consists of the distance from the closest landmark of interest. A Fully-Convolutional Deep Neural Network is optimized to predict this distance for each image location, implicitly casting the problem into a per-pixel Multi-Task Learning approach by which a globally consistent distribution of distances across the entire image can be learned. Once trained, the two minimal distances predicted by the model are selected as the locations of the optic disc and the fovea. The joint learning of every pixel position relative to the optic disc and the fovea favors an automatic understanding of the overall anatomical distribution. This results in an effective technique that can detect both locations simultaneously, as opposed to previous methods that handle both tasks separately. Comprehensive experimental results on a large public dataset validate the proposed approach.
Cornell University - arXiv, Jan 31, 2017
Synthesizing images of the eye fundus is a challenging task that has been previously approached b... more Synthesizing images of the eye fundus is a challenging task that has been previously approached by formulating complex models of the anatomy of the eye. New images can then be generated by sampling a suitable parameter space. In this work, we propose a method that learns to synthesize eye fundus images directly from data. For that, we pair true eye fundus images with their respective vessel trees, by means of a vessel segmentation technique. These pairs are then used to learn a mapping from a binary vessel tree to a new retinal image. For this purpose, we use a recent image-to-image translation technique, based on the idea of adversarial learning. Experimental results show that the original and the generated images are visually different in terms of their global appearance, in spite of sharing the same vessel tree. Additionally, a quantitative quality analysis of the synthetic retinal images confirms that the produced images retain a high proportion of the true image set quality.
Proceedings 10th International Conference on Image Analysis and Processing
In this paper a method for automatic detection of microaneurysms in digital angiograms of the eye... more In this paper a method for automatic detection of microaneurysms in digital angiograms of the eye fundus is described. These lesions of the human retina, a characteristic of the earliest phases of diabetic retinopathy, present themselves in the angiographic images as small, round, hyperfluorescent objects. The proposed method includes initial pre-processing and enhancement steps, followed by object segmentation. In the
Uploads
Papers by Ana Maria Mendonça