We propose a novel BIRADS-SSDL network that integrates clinically-approved breast lesion characte... more We propose a novel BIRADS-SSDL network that integrates clinically-approved breast lesion characteristics (BIRADS features) into a task-oriented Semi-Supervised Deep Learning (SSDL) for accurate diagnosis on ultrasound (US) images with a small training dataset. Breast US images are converted to BIRADS-oriented Feature Maps (BFMs) using a distance-transformation coupled with a Gaussian filter. Then, the converted BFMs are used as the input of an SSDL network, which performs unsupervised Stacked Convolutional Auto-Encoder (SCAE) image reconstruction guided by lesion classification. This integrated multi-task learning allows SCAE to extract image features with the constraints from the lesion classification task, while the lesion classification is achieved by utilizing the SCAE encoder features with a convolutional network. We trained the BIRADS-SSDL network with an alternative learning strategy by balancing reconstruction error and classification label prediction error. To show the effectiveness of our approach, we evaluated it using two breast US image datasets. We compared the performance of the BIRADS-SSDL network with conventional SCAE and SSDL methods that use the original images as inputs, as well as with an SCAE that use BFMs as inputs. Experimental results on two breast US datasets show that BIRADS-SSDL ranked the best among the four networks, with classification accuracy around 94.23±3.33% and 84.38±3.11% on two datasets. In the case of experiments across two datasets collected from two different institution/and US devices, the developed BIRADS-SSDL is generalizable across the different US devices and institutions without overfitting to a single dataset and achieved satisfactory results. Furthermore, we investigate the performance of the proposed method by varying model training strategies, lesion boundary accuracy, and Gaussian filter parameter. Experimental results showed that pre-training strategy can help to speed up model convergence during training but no improvement of classification accuracy on testing dataset. Classification accuracy decreases as segmentation accuracy decreases. The proposed BIRADS-SSDL achieves the best results among the compared methods in each case and has the capacity to deal with multiple different datasets under one model. Compared with state-of-the-art methods, BIRADS-SSDL could be promising for effective breast US computer-aided diagnosis using small datasets.
Efficient, reliable and reproducible target volume delineation is a key step in the effective pla... more Efficient, reliable and reproducible target volume delineation is a key step in the effective planning of breast radiotherapy. However, post-operative breast target delineation is challenging as the contrast between the tumor bed volume (TBV) and normal breast tissue is relatively low in CT images. In this study, we propose to mimic the markerguidance procedure in manual target delineation. We developed a saliency-based deep learning segmentation (SDL-Seg) algorithm for accurate TBV segmentation in post-operative breast irradiation. The SDL-Seg algorithm incorporates saliency information in the form of markers' location cues into a U-Net model. The design forces the model to encode the location-related features, which underscores regions with high saliency levels and suppresses low saliency regions. The saliency maps were generated by identifying markers on CT images. Markers' location were then converted to probability maps using a distance-transformation coupled with a Gaussian filter. Subsequently, the CT images and the corresponding saliency maps formed a multi-channel input for the SDL-Seg network. Our in-house dataset was comprised of 145 prone CT images from 29 post-operative breast cancer patients, who received 5-fraction partial breast irradiation (PBI) regimen on GammaPod. The 29 patients were randomly split into training (19), validation (5) and test (5) sets. The performance of the proposed method was compared against basic U-Net. Our model achieved mean (standard deviation) of 76.4(±2.7) %, 6.76(±1.83) mm, and 1.9(±0.66) mm for DSC, HD95, and ASD respectively on the test set with computation time of below 11 seconds per one CT volume. SDL-Seg showed superior performance relative to basic U-Net for all the evaluation metrics while preserving low computation cost. The findings demonstrate that SDL-Seg is a promising approach for improving the efficiency and accuracy of the online treatment planning procedure of PBI, such as GammaPod based PBI.
Purpose: To develop a method that can extend dose from two isodose surfaces (isosurfaces) to the ... more Purpose: To develop a method that can extend dose from two isodose surfaces (isosurfaces) to the entire patient volume, and to demonstrate its application in radiotherapy plan isodose tuning. Methods: We hypothesized that volumetric dose distribution can be extended from two isosurfaces-the 100% isosurface and a reference isosurface-with the distances to these two surfaces (100 and ref) as extension variables. The extension function is modeled by a threedimensional lookup table (LUT), where voxel dose values from clinical plans are binned by three indexes: 100 , ref and ref (reference dose level). The mean and standard deviation of voxel doses in each bin are calculated and stored in LUT. Volumetric dose extension is performed voxelwisely by indexing the LUT with the 100 , ref and ref of each query voxel. The mean dose stored in the corresponding bin is filled into the query voxel as extended dose, and the standard deviation be filled voxelwisely as the uncertainty of extension result. We applied dose extension in isodose tuning, which aims to tune volumetric dose distribution by isosurface dragging. We adopted extended dose as an approximate dose estimation, and combined it with dose correction strategy to achieve accurate dose tuning. Results: We collected 32 post-operative prostate volumetric modulated arc therapy (VMAT) cases and built the LUT and its associated uncertainties from the doses of 27 cases. The dose extension method was tested on five cases, whose dose distributions were defined as ground truth (GT). We extended the doses from 100% and 50% GT isosurfaces to the entire volume, and evaluated the accuracy of extended doses. The 5mm/5% gamma passing rate (GPR) of extended doses are 92.0%. The mean error is 4.5%, which is consistent to the uncertainty estimated by LUT. The dose difference in 90.5% of voxels is within two sigma and 97.5% in three sigma. The calculation time is less than two seconds. To simulate plan isodose tuning, we optimized a dose with less sparing on rectum (than GT dose) and defined it as a "base dose"-the dose awaiting isosurface dragging. In front-end, the simulated isodose tuning is conducted as such that the base dose was given to plan tuner, and its 50% isosurface would be dragged to the desired position (position of 50% isosurface in GT dose). In back-end, the output of isodose tuning is obtained by 1) extending dose from the desired isosurfaces and viewed the extended dose as an approximate dose, 2) obtaining a correction map from the base dose, and 3) applying the correction map to the extended dose. The accuracy of output-extended dose with correction-was 97.2% in GPR (3mm/3%) and less than 1% in mean dose difference. The total calculation time is less than two seconds, which allows for interactive isodose tuning. Conclusions: We developed a dose extension method that generates volumetric dose distribution from two surfaces. The application of dose extension is in interactive isodose tuning. The distancebased LUT fashion and correction strategy guarantee the computation efficiency and accuracy in isodose tuning. Keywords External beam treatment planning, Dose extension, Isodose tuning Symbols ⃑ 3D point described by Cartesian Coordinates. ⃑ is query point and ⃑ ref is reference point. Surface embedded in 3D Euclidean space. 100% is 100% isodose surface. ref is reference surface. The volume of patient in 3D Euclidean space. 3D dose distribution defined in. ̃ Approximate dose distribution. Extended dose is used as approximate dose in this study. ⃛ Actual dose distribution, which is associated to an existing plan. 100 A scalar field (distance map) that associates the value of ℒ 100 to every query point ⃑ in. ref A scalar field (distance map) that associates the value of ℒ ref to every query point ⃑ in. ref A scalar field (dose map) that associates the value of reference dose to every query point ⃑ in. ℱ A function that extend dose from two surfaces (100% and ref) to volume. ℒ 100 A function that measures the distance between a point and 100% isodose surface. ℒ ref A function that measures the distance between a point and reference surface.
Deep learning–based fluence map prediction (DL-FMP) method has been reported in the literature, w... more Deep learning–based fluence map prediction (DL-FMP) method has been reported in the literature, which generated fluence maps for desired dose by deep neural network (DNN)–based inverse mapping. We hypothesized that DL-FMP is similar to general fluence map optimization (FMO) because it’s theoretically based on a general inverse mapping. We designed four experiments to validate the generalizability of DL-FMP to other types of plans apart from the training data, which contained only clinical head and neck (HN) full-arc volumetric modulated arc therapy (VMAT) plans. The first three experiments quantified the generalizability of DL-FMP to multiple anatomical sites, different delivery modalities, and various degree of modulation (DOM), respectively. The fourth experiment explored the generalizability and stability to infeasible dose inputs. Results of the first experiment manifested that DL-FMP can generalize to lung, liver, esophagus and prostate, with gamma passing rates (GPR) higher th...
Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive perfor... more Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive performance on medical image segmentation. Medical imaging data paired with noisy annotation are, however, ubiquitous, but little is known about the effect of noisy annotation on deep learning-based medical image segmentation. We studied the effects of noisy annotation in the context of mandible segmentation from CT images. First, 202 images of Head and Neck cancer patients were collected from our clinical database, where the organs-at-risk were annotated by one of 12 planning dosimetrists. The mandibles were roughly annotated as the planning avoiding structure. Then, mandible labels were checked and corrected by a physician to get clean annotations. At last, by varying the ratios of noisy labels in the training data, deep learning-based segmentation models were trained, one for each ratio. In general, a deep network trained with noisy labels had worse segmentation results than that trained with clean labels, and fewer noisy labels led to better segmentation. When using 20% or less noisy cases for training, no significant difference was found on the prediction performance between the models trained by noisy or clean. This study suggests that deep learning-based medical image segmentation is robust to noisy annotations to some extent. It also highlights the importance of labeling quality in deep learning.
A novel method was developed to track lung tumor motion in real time during radiation therapy wit... more A novel method was developed to track lung tumor motion in real time during radiation therapy with the purpose to allow target radiation dose escalation while simultaneously reducing the dose to sensitive structures, thereby increasing local control without increasing toxicity. This method analyzes beam’s eye view radiation therapy treatment megavoltage (MV) images with simulated digitally reconstructed radiographs (DRRs) as references. Instead of comparing global DRRs with projection images, this method incorporates a technique that divides the global composite DRR and the corresponding MV projection into sub-images called tiles. Registration is performed independently on tile pairs in order to reduce the effects of global discrepancies due to scattering or imaging modality differences. This algorithm was evaluated by phantom studies while simulated tumors were controlled to move with various patterns in a complex humanoid torso. Approximately 15,000 phantom MV images were acquired...
ABSTRACT Purpose: During a typical 5-7 week treatment of external beam radiotherapy, there are po... more ABSTRACT Purpose: During a typical 5-7 week treatment of external beam radiotherapy, there are potential differences between planned patient's anatomy and positioning, such as patient weight loss, or treatment setup. The discrepancies between planned and delivered doses resulting from these differences could be significant, especially in IMRT where dose distributions tightly conforms to target volumes while avoiding organs-at-risk. We developed an automatic system to monitor delivered dose using daily imaging. Methods: For each treatment, a merged image is generated by registering the daily pre-treatment setup image and planning CT using treatment position information extracted from the Tomotherapy archive. The treatment dose is then computed on this merged image using our in-house convolution-superposition based dose calculator implemented on GPU. The deformation field between merged and planning CT is computed using the Morphon algorithm. The planning structures and treatment doses are subsequently warped for analysis and dose accumulation. All results are saved in DICOM format with private tags and organized in a database. Due to the overwhelming amount of information generated, a customizable tolerance system is used to flag potential treatment errors or significant anatomical changes. A web-based system and a DICOM-RT viewer were developed for reporting and reviewing the results. Results: More than 30 patients were analysed retrospectively. Our in-house dose calculator passed 97% gamma test evaluated with 2% dose difference and 2mm distance-to-agreement compared with Tomotherapy calculated dose, which is considered sufficient for adaptive radiotherapy purposes. Evaluation of the deformable registration through visual inspection showed acceptable and consistent results, except for cases with large or unrealistic deformation. Our automatic flagging system was able to catch significant patient setup errors or anatomical changes. Conclusions: We developed an automatic dose verification system that quantifies treatment doses, and provides necessary information for adaptive planning without impeding clinical workflows.
Adaptive radiotherapy (ART), especially online ART, effectively accounts for positioning errors a... more Adaptive radiotherapy (ART), especially online ART, effectively accounts for positioning errors and anatomical changes during the course of treatment. One key component of online ART processes is accurately and efficiently delineating organs at risk (OARs) and targets on online images, such as Cone Beam Computed Tomography (CBCT), to meet the online demands of plan evaluation and adaptation. Deep learning (DL)-based automatic segmentation has gained great success in segmenting planning CT, but its applications to CBCT yielded inferior results due to the low image quality and limited available contour labels for training. To overcome these obstacles to online CBCT segmentation, we propose a registration-guided DL (RgDL) segmentation framework that integrates image registration algorithms and DL segmentation models. The registration algorithm generates initial contours, which were used as guidance by DL model to obtain the accurate final segmentations. We had two implementations the proposed framework-Rig-RgDL (Rig for rigid body) and Def-RgDL (Def for deformable)-with rigid body (RB) registration or deformable image registration (DIR) as the registration algorithm respectively and U-Net as DL model architecture. The two implementations of RgDL framework were trained and evaluated on seven OARs in an institutional clinical Head and Neck (HN) dataset. Compared to the baseline approaches using the registration or the DL alone, RgDL achieved more accurate segmentation, as measured by higher mean Dice similarity coefficients (DSC) and other distance-based metrics. Rig-RgDL achieved a DSC of 84.5% on seven OARs on average, higher than RB or DL alone by 4.5% and 4.7%. The DSC of Def-RgDL is 86.5%, higher than DIR or DL alone by 2.4% and 6.7%. The inference time took by the DL model to generate final segmentations of seven OARs is less than one second in RgDL. The resulting segmentation accuracy and efficiency show the promise of applying RgDL framework for online ART.
Deep learning has started to revolutionize several different industries, and the applications of ... more Deep learning has started to revolutionize several different industries, and the applications of these methods in medicine are now becoming more commonplace. This study focuses on investigating the feasibility of tracking patients and clinical staff wearing Bluetooth Low Energy (BLE) tags in a radiation oncology clinic using artificial neural networks (ANNs) and convolutional neural networks (CNNs). The performance of these networks was compared to relative received signal strength indicator (RSSI) thresholding and triangulation. By utilizing temporal information, a combined CNN+ANN network was capable of correctly identifying the location of the BLE tag with an accuracy of 99.9%. It outperformed a CNN model (accuracy = 94%), a thresholding model employing majority voting (accuracy = 95%), and a triangulation classifier utilizing majority voting (accuracy = 95%). Future studies will seek to deploy this affordable real time location system in hospitals to improve clinical workflow, e...
We study threshold-driven optimization methodology for automatically generating a treatment plan ... more We study threshold-driven optimization methodology for automatically generating a treatment plan that is motivated by a reference DVH for IMRT treatment planning. We present a framework for threshold-driven optimization for reference-based auto-planning (TORA). Commonly used voxel-based quadratic penalties have two components for penalizing under- and over-dosing of voxels: a reference dose threshold and associated penalty weight. Conventional manual- and auto-planning using such a function involves iteratively updating the preference weights while keeping the thresholds constant, an unintuitive and often inconsistent method for planning toward some reference DVH. However, driving a dose distribution by threshold values instead of preference weights can achieve similar plans with less computational effort. The proposed methodology spatially assigns reference DVH information to threshold values, and iteratively improves the quality of that assignment. The methodology effectively hand...
Journal of applied clinical medical physics, Jan 3, 2018
Electron therapy is widely used to treat shallow tumors because of its characteristic sharp dose ... more Electron therapy is widely used to treat shallow tumors because of its characteristic sharp dose fall-off beyond a certain range. A customized cutout is typically applied to block radiation to normal tissues. Determining the final monitor unit (MU) for electron treatment requires an output factor for the cutout, which is usually generated by measurement, especially for highly irregular cutouts. However, manual measurement requires a lengthy quality assurance process with possible errors. This work presents an accurate and efficient cutout output factor prediction model, convolution-based modified Clarkson integration (CMCI), to replace patient-specific output factor measurement. Like the Clarkson method, we decompose the field into basic sectors. Unlike the Clarkson integration method, we use annular sectors for output factor estimation. This decomposition method allows calculation via convolution. A 2D distribution of fluence is generated, and the output factor at any given point c...
Accurate and automatic brain metastases target delineation is a key step for efficient and effect... more Accurate and automatic brain metastases target delineation is a key step for efficient and effective stereotactic radiosurgery (SRS) treatment planning. In this work, we developed a deep learning convolutional neural network (CNN) algorithm for segmenting brain metastases on contrast-enhanced T1-weighted magnetic resonance imaging (MRI) datasets. We integrated the CNN-based algorithm into an automatic brain metastases segmentation workflow and validated on both Multimodal Brain Tumor Image Segmentation challenge (BRATS) data and clinical patients' data. Validation on BRATS data yielded average DICE coefficients (DCs) of 0.75±0.07 in the tumor core and 0.81±0.04 in the enhancing tumor, which outperformed most techniques in the 2015 BRATS challenge. Segmentation results of patient cases showed an average of DCs 0.67±0.03 and achieved an area under the receiver operating characteristic curve of 0.98±0.01. The developed automatic segmentation strategy surpasses current benchmark lev...
Conventional step-and-shoot intensity modulated radiation therapy leaf sequencing methods, where ... more Conventional step-and-shoot intensity modulated radiation therapy leaf sequencing methods, where a nonhomogeneous fluence map is converted to a set of apertures and associated intensities, assume that target fluence is stratified into a fixed number of discrete levels and/or aperture leaf positions are restricted to a discrete set of locations. These assumptions induce a deviation from the planned fluence map and/or reduce the feasible region of potential plans, respectively. A continuous leaf optimization (CLO) framework is developed as a postprocessing methodology to improve upon conventional leaf sequencing so that the resulting plan avoids these two main drawbacks. The CLO model directly represents leaf positions and aperture intensities with continuous variables with the goal of reproducing some target fluence profile. Fluence through leaf edges is modeled using the error function, and continuous fluence is approximated using a 0.1 mm discretization across the aperture opening....
Multi-atlas segmentation (MAS) has been widely used to automate the delineation of organs at risk... more Multi-atlas segmentation (MAS) has been widely used to automate the delineation of organs at risk (OARs) for radiotherapy. Label fusion is a crucial step in MAS to cope with the segmentation variabilities among multiple atlases. However, most existing label fusion methods do not consider the potential dosimetric impact of the segmentation result. In this proof-of-concept study, we propose a novel geometry-dosimetry label fusion method for MAS-based OAR auto-contouring, which evaluates the segmentation performance in terms of both geometric accuracy and the dosimetric impact of the segmentation accuracy on the resulting treatment plan. Differently from the original selective and iterative method for performance level estimation (SIMPLE), we evaluated and rejected the atlases based on both Dice similarity coefficient and the predicted error of the dosimetric endpoints. The dosimetric error was predicted using our previously developed geometry-dosimetry model. We tested our method in M...
We have initiated a multi-institutional phase I trial of 5-fraction stereotactic body radiotherap... more We have initiated a multi-institutional phase I trial of 5-fraction stereotactic body radiotherapy (SBRT) for Stage III-IVa laryngeal cancer. We conducted this pilot dosimetric study to confirm potential utility of online adaptive replanning to preserve treatment quality. We evaluated ten cases: five patients enrolled onto the current trial and five patients enrolled onto a separate phase I SBRT trial for early-stage glottic larynx cancer. Baseline SBRT treatment plans were generated per protocol. Daily cone-beam CT (CBCT) or diagnostic CT images were acquired prior to each treatment fraction. Simulation CT images and target volumes were deformably registered to daily volumetric images, the original SBRT plan was copied to the deformed images and contours, delivered dose distributions were re-calculated on the deformed CT images. All of these were performed on a commercial treatment planning system. In-house software was developed to propagate the delivered dose distribution back to...
We propose a novel BIRADS-SSDL network that integrates clinically-approved breast lesion characte... more We propose a novel BIRADS-SSDL network that integrates clinically-approved breast lesion characteristics (BIRADS features) into a task-oriented Semi-Supervised Deep Learning (SSDL) for accurate diagnosis on ultrasound (US) images with a small training dataset. Breast US images are converted to BIRADS-oriented Feature Maps (BFMs) using a distance-transformation coupled with a Gaussian filter. Then, the converted BFMs are used as the input of an SSDL network, which performs unsupervised Stacked Convolutional Auto-Encoder (SCAE) image reconstruction guided by lesion classification. This integrated multi-task learning allows SCAE to extract image features with the constraints from the lesion classification task, while the lesion classification is achieved by utilizing the SCAE encoder features with a convolutional network. We trained the BIRADS-SSDL network with an alternative learning strategy by balancing reconstruction error and classification label prediction error. To show the effectiveness of our approach, we evaluated it using two breast US image datasets. We compared the performance of the BIRADS-SSDL network with conventional SCAE and SSDL methods that use the original images as inputs, as well as with an SCAE that use BFMs as inputs. Experimental results on two breast US datasets show that BIRADS-SSDL ranked the best among the four networks, with classification accuracy around 94.23±3.33% and 84.38±3.11% on two datasets. In the case of experiments across two datasets collected from two different institution/and US devices, the developed BIRADS-SSDL is generalizable across the different US devices and institutions without overfitting to a single dataset and achieved satisfactory results. Furthermore, we investigate the performance of the proposed method by varying model training strategies, lesion boundary accuracy, and Gaussian filter parameter. Experimental results showed that pre-training strategy can help to speed up model convergence during training but no improvement of classification accuracy on testing dataset. Classification accuracy decreases as segmentation accuracy decreases. The proposed BIRADS-SSDL achieves the best results among the compared methods in each case and has the capacity to deal with multiple different datasets under one model. Compared with state-of-the-art methods, BIRADS-SSDL could be promising for effective breast US computer-aided diagnosis using small datasets.
Efficient, reliable and reproducible target volume delineation is a key step in the effective pla... more Efficient, reliable and reproducible target volume delineation is a key step in the effective planning of breast radiotherapy. However, post-operative breast target delineation is challenging as the contrast between the tumor bed volume (TBV) and normal breast tissue is relatively low in CT images. In this study, we propose to mimic the markerguidance procedure in manual target delineation. We developed a saliency-based deep learning segmentation (SDL-Seg) algorithm for accurate TBV segmentation in post-operative breast irradiation. The SDL-Seg algorithm incorporates saliency information in the form of markers' location cues into a U-Net model. The design forces the model to encode the location-related features, which underscores regions with high saliency levels and suppresses low saliency regions. The saliency maps were generated by identifying markers on CT images. Markers' location were then converted to probability maps using a distance-transformation coupled with a Gaussian filter. Subsequently, the CT images and the corresponding saliency maps formed a multi-channel input for the SDL-Seg network. Our in-house dataset was comprised of 145 prone CT images from 29 post-operative breast cancer patients, who received 5-fraction partial breast irradiation (PBI) regimen on GammaPod. The 29 patients were randomly split into training (19), validation (5) and test (5) sets. The performance of the proposed method was compared against basic U-Net. Our model achieved mean (standard deviation) of 76.4(±2.7) %, 6.76(±1.83) mm, and 1.9(±0.66) mm for DSC, HD95, and ASD respectively on the test set with computation time of below 11 seconds per one CT volume. SDL-Seg showed superior performance relative to basic U-Net for all the evaluation metrics while preserving low computation cost. The findings demonstrate that SDL-Seg is a promising approach for improving the efficiency and accuracy of the online treatment planning procedure of PBI, such as GammaPod based PBI.
Purpose: To develop a method that can extend dose from two isodose surfaces (isosurfaces) to the ... more Purpose: To develop a method that can extend dose from two isodose surfaces (isosurfaces) to the entire patient volume, and to demonstrate its application in radiotherapy plan isodose tuning. Methods: We hypothesized that volumetric dose distribution can be extended from two isosurfaces-the 100% isosurface and a reference isosurface-with the distances to these two surfaces (100 and ref) as extension variables. The extension function is modeled by a threedimensional lookup table (LUT), where voxel dose values from clinical plans are binned by three indexes: 100 , ref and ref (reference dose level). The mean and standard deviation of voxel doses in each bin are calculated and stored in LUT. Volumetric dose extension is performed voxelwisely by indexing the LUT with the 100 , ref and ref of each query voxel. The mean dose stored in the corresponding bin is filled into the query voxel as extended dose, and the standard deviation be filled voxelwisely as the uncertainty of extension result. We applied dose extension in isodose tuning, which aims to tune volumetric dose distribution by isosurface dragging. We adopted extended dose as an approximate dose estimation, and combined it with dose correction strategy to achieve accurate dose tuning. Results: We collected 32 post-operative prostate volumetric modulated arc therapy (VMAT) cases and built the LUT and its associated uncertainties from the doses of 27 cases. The dose extension method was tested on five cases, whose dose distributions were defined as ground truth (GT). We extended the doses from 100% and 50% GT isosurfaces to the entire volume, and evaluated the accuracy of extended doses. The 5mm/5% gamma passing rate (GPR) of extended doses are 92.0%. The mean error is 4.5%, which is consistent to the uncertainty estimated by LUT. The dose difference in 90.5% of voxels is within two sigma and 97.5% in three sigma. The calculation time is less than two seconds. To simulate plan isodose tuning, we optimized a dose with less sparing on rectum (than GT dose) and defined it as a "base dose"-the dose awaiting isosurface dragging. In front-end, the simulated isodose tuning is conducted as such that the base dose was given to plan tuner, and its 50% isosurface would be dragged to the desired position (position of 50% isosurface in GT dose). In back-end, the output of isodose tuning is obtained by 1) extending dose from the desired isosurfaces and viewed the extended dose as an approximate dose, 2) obtaining a correction map from the base dose, and 3) applying the correction map to the extended dose. The accuracy of output-extended dose with correction-was 97.2% in GPR (3mm/3%) and less than 1% in mean dose difference. The total calculation time is less than two seconds, which allows for interactive isodose tuning. Conclusions: We developed a dose extension method that generates volumetric dose distribution from two surfaces. The application of dose extension is in interactive isodose tuning. The distancebased LUT fashion and correction strategy guarantee the computation efficiency and accuracy in isodose tuning. Keywords External beam treatment planning, Dose extension, Isodose tuning Symbols ⃑ 3D point described by Cartesian Coordinates. ⃑ is query point and ⃑ ref is reference point. Surface embedded in 3D Euclidean space. 100% is 100% isodose surface. ref is reference surface. The volume of patient in 3D Euclidean space. 3D dose distribution defined in. ̃ Approximate dose distribution. Extended dose is used as approximate dose in this study. ⃛ Actual dose distribution, which is associated to an existing plan. 100 A scalar field (distance map) that associates the value of ℒ 100 to every query point ⃑ in. ref A scalar field (distance map) that associates the value of ℒ ref to every query point ⃑ in. ref A scalar field (dose map) that associates the value of reference dose to every query point ⃑ in. ℱ A function that extend dose from two surfaces (100% and ref) to volume. ℒ 100 A function that measures the distance between a point and 100% isodose surface. ℒ ref A function that measures the distance between a point and reference surface.
Deep learning–based fluence map prediction (DL-FMP) method has been reported in the literature, w... more Deep learning–based fluence map prediction (DL-FMP) method has been reported in the literature, which generated fluence maps for desired dose by deep neural network (DNN)–based inverse mapping. We hypothesized that DL-FMP is similar to general fluence map optimization (FMO) because it’s theoretically based on a general inverse mapping. We designed four experiments to validate the generalizability of DL-FMP to other types of plans apart from the training data, which contained only clinical head and neck (HN) full-arc volumetric modulated arc therapy (VMAT) plans. The first three experiments quantified the generalizability of DL-FMP to multiple anatomical sites, different delivery modalities, and various degree of modulation (DOM), respectively. The fourth experiment explored the generalizability and stability to infeasible dose inputs. Results of the first experiment manifested that DL-FMP can generalize to lung, liver, esophagus and prostate, with gamma passing rates (GPR) higher th...
Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive perfor... more Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive performance on medical image segmentation. Medical imaging data paired with noisy annotation are, however, ubiquitous, but little is known about the effect of noisy annotation on deep learning-based medical image segmentation. We studied the effects of noisy annotation in the context of mandible segmentation from CT images. First, 202 images of Head and Neck cancer patients were collected from our clinical database, where the organs-at-risk were annotated by one of 12 planning dosimetrists. The mandibles were roughly annotated as the planning avoiding structure. Then, mandible labels were checked and corrected by a physician to get clean annotations. At last, by varying the ratios of noisy labels in the training data, deep learning-based segmentation models were trained, one for each ratio. In general, a deep network trained with noisy labels had worse segmentation results than that trained with clean labels, and fewer noisy labels led to better segmentation. When using 20% or less noisy cases for training, no significant difference was found on the prediction performance between the models trained by noisy or clean. This study suggests that deep learning-based medical image segmentation is robust to noisy annotations to some extent. It also highlights the importance of labeling quality in deep learning.
A novel method was developed to track lung tumor motion in real time during radiation therapy wit... more A novel method was developed to track lung tumor motion in real time during radiation therapy with the purpose to allow target radiation dose escalation while simultaneously reducing the dose to sensitive structures, thereby increasing local control without increasing toxicity. This method analyzes beam’s eye view radiation therapy treatment megavoltage (MV) images with simulated digitally reconstructed radiographs (DRRs) as references. Instead of comparing global DRRs with projection images, this method incorporates a technique that divides the global composite DRR and the corresponding MV projection into sub-images called tiles. Registration is performed independently on tile pairs in order to reduce the effects of global discrepancies due to scattering or imaging modality differences. This algorithm was evaluated by phantom studies while simulated tumors were controlled to move with various patterns in a complex humanoid torso. Approximately 15,000 phantom MV images were acquired...
ABSTRACT Purpose: During a typical 5-7 week treatment of external beam radiotherapy, there are po... more ABSTRACT Purpose: During a typical 5-7 week treatment of external beam radiotherapy, there are potential differences between planned patient's anatomy and positioning, such as patient weight loss, or treatment setup. The discrepancies between planned and delivered doses resulting from these differences could be significant, especially in IMRT where dose distributions tightly conforms to target volumes while avoiding organs-at-risk. We developed an automatic system to monitor delivered dose using daily imaging. Methods: For each treatment, a merged image is generated by registering the daily pre-treatment setup image and planning CT using treatment position information extracted from the Tomotherapy archive. The treatment dose is then computed on this merged image using our in-house convolution-superposition based dose calculator implemented on GPU. The deformation field between merged and planning CT is computed using the Morphon algorithm. The planning structures and treatment doses are subsequently warped for analysis and dose accumulation. All results are saved in DICOM format with private tags and organized in a database. Due to the overwhelming amount of information generated, a customizable tolerance system is used to flag potential treatment errors or significant anatomical changes. A web-based system and a DICOM-RT viewer were developed for reporting and reviewing the results. Results: More than 30 patients were analysed retrospectively. Our in-house dose calculator passed 97% gamma test evaluated with 2% dose difference and 2mm distance-to-agreement compared with Tomotherapy calculated dose, which is considered sufficient for adaptive radiotherapy purposes. Evaluation of the deformable registration through visual inspection showed acceptable and consistent results, except for cases with large or unrealistic deformation. Our automatic flagging system was able to catch significant patient setup errors or anatomical changes. Conclusions: We developed an automatic dose verification system that quantifies treatment doses, and provides necessary information for adaptive planning without impeding clinical workflows.
Adaptive radiotherapy (ART), especially online ART, effectively accounts for positioning errors a... more Adaptive radiotherapy (ART), especially online ART, effectively accounts for positioning errors and anatomical changes during the course of treatment. One key component of online ART processes is accurately and efficiently delineating organs at risk (OARs) and targets on online images, such as Cone Beam Computed Tomography (CBCT), to meet the online demands of plan evaluation and adaptation. Deep learning (DL)-based automatic segmentation has gained great success in segmenting planning CT, but its applications to CBCT yielded inferior results due to the low image quality and limited available contour labels for training. To overcome these obstacles to online CBCT segmentation, we propose a registration-guided DL (RgDL) segmentation framework that integrates image registration algorithms and DL segmentation models. The registration algorithm generates initial contours, which were used as guidance by DL model to obtain the accurate final segmentations. We had two implementations the proposed framework-Rig-RgDL (Rig for rigid body) and Def-RgDL (Def for deformable)-with rigid body (RB) registration or deformable image registration (DIR) as the registration algorithm respectively and U-Net as DL model architecture. The two implementations of RgDL framework were trained and evaluated on seven OARs in an institutional clinical Head and Neck (HN) dataset. Compared to the baseline approaches using the registration or the DL alone, RgDL achieved more accurate segmentation, as measured by higher mean Dice similarity coefficients (DSC) and other distance-based metrics. Rig-RgDL achieved a DSC of 84.5% on seven OARs on average, higher than RB or DL alone by 4.5% and 4.7%. The DSC of Def-RgDL is 86.5%, higher than DIR or DL alone by 2.4% and 6.7%. The inference time took by the DL model to generate final segmentations of seven OARs is less than one second in RgDL. The resulting segmentation accuracy and efficiency show the promise of applying RgDL framework for online ART.
Deep learning has started to revolutionize several different industries, and the applications of ... more Deep learning has started to revolutionize several different industries, and the applications of these methods in medicine are now becoming more commonplace. This study focuses on investigating the feasibility of tracking patients and clinical staff wearing Bluetooth Low Energy (BLE) tags in a radiation oncology clinic using artificial neural networks (ANNs) and convolutional neural networks (CNNs). The performance of these networks was compared to relative received signal strength indicator (RSSI) thresholding and triangulation. By utilizing temporal information, a combined CNN+ANN network was capable of correctly identifying the location of the BLE tag with an accuracy of 99.9%. It outperformed a CNN model (accuracy = 94%), a thresholding model employing majority voting (accuracy = 95%), and a triangulation classifier utilizing majority voting (accuracy = 95%). Future studies will seek to deploy this affordable real time location system in hospitals to improve clinical workflow, e...
We study threshold-driven optimization methodology for automatically generating a treatment plan ... more We study threshold-driven optimization methodology for automatically generating a treatment plan that is motivated by a reference DVH for IMRT treatment planning. We present a framework for threshold-driven optimization for reference-based auto-planning (TORA). Commonly used voxel-based quadratic penalties have two components for penalizing under- and over-dosing of voxels: a reference dose threshold and associated penalty weight. Conventional manual- and auto-planning using such a function involves iteratively updating the preference weights while keeping the thresholds constant, an unintuitive and often inconsistent method for planning toward some reference DVH. However, driving a dose distribution by threshold values instead of preference weights can achieve similar plans with less computational effort. The proposed methodology spatially assigns reference DVH information to threshold values, and iteratively improves the quality of that assignment. The methodology effectively hand...
Journal of applied clinical medical physics, Jan 3, 2018
Electron therapy is widely used to treat shallow tumors because of its characteristic sharp dose ... more Electron therapy is widely used to treat shallow tumors because of its characteristic sharp dose fall-off beyond a certain range. A customized cutout is typically applied to block radiation to normal tissues. Determining the final monitor unit (MU) for electron treatment requires an output factor for the cutout, which is usually generated by measurement, especially for highly irregular cutouts. However, manual measurement requires a lengthy quality assurance process with possible errors. This work presents an accurate and efficient cutout output factor prediction model, convolution-based modified Clarkson integration (CMCI), to replace patient-specific output factor measurement. Like the Clarkson method, we decompose the field into basic sectors. Unlike the Clarkson integration method, we use annular sectors for output factor estimation. This decomposition method allows calculation via convolution. A 2D distribution of fluence is generated, and the output factor at any given point c...
Accurate and automatic brain metastases target delineation is a key step for efficient and effect... more Accurate and automatic brain metastases target delineation is a key step for efficient and effective stereotactic radiosurgery (SRS) treatment planning. In this work, we developed a deep learning convolutional neural network (CNN) algorithm for segmenting brain metastases on contrast-enhanced T1-weighted magnetic resonance imaging (MRI) datasets. We integrated the CNN-based algorithm into an automatic brain metastases segmentation workflow and validated on both Multimodal Brain Tumor Image Segmentation challenge (BRATS) data and clinical patients' data. Validation on BRATS data yielded average DICE coefficients (DCs) of 0.75±0.07 in the tumor core and 0.81±0.04 in the enhancing tumor, which outperformed most techniques in the 2015 BRATS challenge. Segmentation results of patient cases showed an average of DCs 0.67±0.03 and achieved an area under the receiver operating characteristic curve of 0.98±0.01. The developed automatic segmentation strategy surpasses current benchmark lev...
Conventional step-and-shoot intensity modulated radiation therapy leaf sequencing methods, where ... more Conventional step-and-shoot intensity modulated radiation therapy leaf sequencing methods, where a nonhomogeneous fluence map is converted to a set of apertures and associated intensities, assume that target fluence is stratified into a fixed number of discrete levels and/or aperture leaf positions are restricted to a discrete set of locations. These assumptions induce a deviation from the planned fluence map and/or reduce the feasible region of potential plans, respectively. A continuous leaf optimization (CLO) framework is developed as a postprocessing methodology to improve upon conventional leaf sequencing so that the resulting plan avoids these two main drawbacks. The CLO model directly represents leaf positions and aperture intensities with continuous variables with the goal of reproducing some target fluence profile. Fluence through leaf edges is modeled using the error function, and continuous fluence is approximated using a 0.1 mm discretization across the aperture opening....
Multi-atlas segmentation (MAS) has been widely used to automate the delineation of organs at risk... more Multi-atlas segmentation (MAS) has been widely used to automate the delineation of organs at risk (OARs) for radiotherapy. Label fusion is a crucial step in MAS to cope with the segmentation variabilities among multiple atlases. However, most existing label fusion methods do not consider the potential dosimetric impact of the segmentation result. In this proof-of-concept study, we propose a novel geometry-dosimetry label fusion method for MAS-based OAR auto-contouring, which evaluates the segmentation performance in terms of both geometric accuracy and the dosimetric impact of the segmentation accuracy on the resulting treatment plan. Differently from the original selective and iterative method for performance level estimation (SIMPLE), we evaluated and rejected the atlases based on both Dice similarity coefficient and the predicted error of the dosimetric endpoints. The dosimetric error was predicted using our previously developed geometry-dosimetry model. We tested our method in M...
We have initiated a multi-institutional phase I trial of 5-fraction stereotactic body radiotherap... more We have initiated a multi-institutional phase I trial of 5-fraction stereotactic body radiotherapy (SBRT) for Stage III-IVa laryngeal cancer. We conducted this pilot dosimetric study to confirm potential utility of online adaptive replanning to preserve treatment quality. We evaluated ten cases: five patients enrolled onto the current trial and five patients enrolled onto a separate phase I SBRT trial for early-stage glottic larynx cancer. Baseline SBRT treatment plans were generated per protocol. Daily cone-beam CT (CBCT) or diagnostic CT images were acquired prior to each treatment fraction. Simulation CT images and target volumes were deformably registered to daily volumetric images, the original SBRT plan was copied to the deformed images and contours, delivered dose distributions were re-calculated on the deformed CT images. All of these were performed on a commercial treatment planning system. In-house software was developed to propagate the delivered dose distribution back to...
Uploads
Papers by Weiguo Lu