bioRxiv (Cold Spring Harbor Laboratory), Dec 6, 2023
This study presents an innovative approach for understanding the genetic underpinnings of two key... more This study presents an innovative approach for understanding the genetic underpinnings of two key phenotypes in Sorghum bicolor: maximum canopy height and maximum growth rate. Genome-Wide Association Studies (GWAS) are widely used to decipher the genetic basis of traits in organisms, but the challenge lies in selecting an appropriate statistically significant threshold for analysis. Our goal was to employ GWAS to pinpoint the genetic markers associated with the phenotypes of interest using specific permissive-filtered threshold values that allows the inclusion of broader collections of explanatory candidate genes. Then, we utilized a pattern recognition technique to prioritize a set of informative genes, which hold potential for further investigation and could find applications in Artificial Intelligence systems. Utilizing a subset of the Sorghum Bioenergy Association Panel cultivated at the Maricopa Agricultural Center in Arizona, we sought to unveil patterns between phenotypic similarity and genetic proximity among accessions in order to organize Single Nucleotide Polymorphisms (SNPs) which are likely to be associated with the phenotypic trait. Additionally, we explored the impact of this method by considering all SNPs versus focusing on SNPs classified through the GWAS pre-filter. Experimental results indicated that our approach effectively prioritizes SNPs and genes influencing the phenotype of interest. Moreover, this methodology holds promise in the feature selection from genomic data for predicting complex phenotypic traits influenced by numerous genes and environmental conditions and could pave the way for further research in this field.
Members of different species often engage in aggressive contests over resources. This series of a... more Members of different species often engage in aggressive contests over resources. This series of aggressive contests between species may result in an interspecific dominance hierarchy. Such hierarchies are of interest because they could be used to address a variety of research questions, for example, do similarly ranked species tend to avoid each other in time or space, and what will happen when such species come into contact as climates change? Here, we propose a method for creating a continental-scale hierarchy, and we make initial analyses based on this hierarchy. Leveraging the existing network of citizen scientists from Project FeederWatch, we collected the data with which to create a continent-spanning interspecific dominance hierarchy that included species that do not currently have overlapping geographic distributions. We quantified the extent of intransitivities (rock-paper-scissors relationships) in the hierarchy, as intransitivities can promote local species’ coexistence. Overall, the hierarchy was nearly linear, and largely predicted by body mass, although there were clade-specific deviations from the average mass–dominance relationship. Warblers and orioles, for instance, were more dominant than expected based on their body mass, while buntings, grosbeaks, and doves were less dominant than expected. Intransitive relationships were rare. Few interactions were reported between close relatives and ecological competitors like Mountain and Black-capped Chickadees, as such species often have only marginally overlapping geographic distributions, restricting opportunity for observation. Yet, these species’ ranks—emergent properties of the network—were often in agreement with targeted studies of dominance relationships between them
The study of aggressive interactions between species has, to date, usually been restricted to int... more The study of aggressive interactions between species has, to date, usually been restricted to interactions among small numbers of ecologically close competitors. Nothing is known about interspecific dominance hierarchies that include numerous, ecologically varied species. Such hierarchies are of interest because they could be used to address a variety of research questions, e.g. do similarly ranked species tend to avoid each other in time or Interspecific dominance hierarchy, aggression, displacement, citizen science LAY SUMMARY When it comes to fighting over food, bigger is better but woodpeckers are best. The outcome of aggressive encounters between birds frequently determines which individual gains access to contested resources like food, but until now, little was known about such encounters between individuals of different species. We partnered with citizen scientists to record interspecific behavioral interactions at bird feeders around North America, and assembled these interactions into a continental dominance hierarchy. .
Supplemental Table S1. Whole slide image tissue segmentation statistics by an expert pathologist ... more Supplemental Table S1. Whole slide image tissue segmentation statistics by an expert pathologist and probability prediction using a trained convolutional neural network. Supplemental Table S2. Clinical and molecular characteristics of FN-RMS samples used for training models for mutation prediction. Yellow boxes indicate genes included in defining the RAS pathway. Supplemental Table S3. Clinical information with COG risk stratification of FN-RMS samples used for training a prognostication predictive CNN.
A.I. classification of duplicate cores Match Mismatch Total Number of patients 109 13 122 Proport... more A.I. classification of duplicate cores Match Mismatch Total Number of patients 109 13 122 Proportion (95% C.
Supplemental Figure S6. Graphical User Interface for ssue segmenta on, MYOD1 muta on predic on, a... more Supplemental Figure S6. Graphical User Interface for ssue segmenta on, MYOD1 muta on predic on, and risk predic on models. (a) A containerized docker was developed to allow users to upload H&E images of RMS ssue and perform ssue segmenta on to classify regions of tumor. (b) Propor ons of ssue predicted as ARMS, ERMS, necrosis, and stroma are generated as a bar graph. (c) User-provided images can also be analyzed for MYOD1 muta ons using our trained MYOD1 muta on predic on model. The posi ve predic on score can be compared with our training cohort of samples that have known MYOD1 muta on status. (d) A hazard predic on score (-1 to +1) can be a ained using the trained FN-RMS risk predic on model. The hazard predic on score can be compared with A.I. predicted risk grouping for our study cohort data (n=264).
Purpose: Rhabdomyosarcoma (RMS) is an aggressive softtissue sarcoma, which primarily occurs in ch... more Purpose: Rhabdomyosarcoma (RMS) is an aggressive softtissue sarcoma, which primarily occurs in children and young adults. We previously reported specific genomic alterations in RMS, which strongly correlated with survival; however, predicting these mutations or high-risk disease at diagnosis remains a significant challenge. In this study, we utilized convolutional neural networks (CNN) to learn histologic features associated with driver mutations and outcome using hematoxylin and eosin (H&E) images of RMS. Experimental Design: Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from 321 patients with RMS enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n ¼ 136) or holdout test data. Results: The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared with current molecular-clinical risk stratification. Conclusions: This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a powerful tool for diagnostic and prognostic prediction of rhabdomyosarcoma, which will be tested in prospective COG clinical trials.
Abstract This chapter is designed to help the reader become familiar with standard methods for de... more Abstract This chapter is designed to help the reader become familiar with standard methods for describing pelvic organ geometry, suitable for biomechanical computational analysis. To perform structural engineering analysis of the pelvic floor tissues, they must first be represented geometrically. This is done via a technique called segmentation, in which individual organs and/or tissue elements are outlined, and given unique labels. The interesting tissues are usually identified on standard radiologic image stacks — usually magnetic resonance imaging (MRI) or computed tomography (CT) scans. The goal of segmentation is to identify a desired tissue on the original (grayscale) image, and trace out (or label) its boundaries. For each organ or tissue of interest, the labeling is performed, either manually or semiautomatically. Because pelvic floor structures exist in a three-dimensional (3-D) space, it is necessary to represent the tissues as 3-D structures. This 3-D representation is facilitated by the method of acquiring MRI and CT image data. Both CT and MRI will produce grayscale source image data sets. Each data set consists of a “stack” of 2-D images, covering the region of the body that was scanned. The extent of the scanned region is called the field of view, and the grayscale images are sometimes referred to as the source images. This chapter will discuss the geometric representation of pelvic floor MRI grayscale data, as well as the segmentation of this data to produce labelmaps suitable for computational analysis and 3-D reconstruction of the organs and tissues of interest. Manual, semiautomatic, and automatic segmentation methods will be briefly introduced, in preparation for a more detailed discussion of automated segmentation in a later chapter. The chapter will conclude with a review of the trade-offs between the segmentation methods, and discussion of the issues with repeatability, reliability, and fidelity of these methods. The chapter focuses on MRI source data, with the understanding that the general principles will apply to CT sourced images as well.
In this paper, we present a molecular modeling approach based on recursive ot~ect-oriented class ... more In this paper, we present a molecular modeling approach based on recursive ot~ect-oriented class instances. Our approach allows ease of use by the scientist and includes a heuristic for evaluating molecular interaction and rendering at multiple levels of detail. Our implementation in C++ employs abstractions to encapsulate the implementation and details of the computations. Visualization is accomplished with OpenGL or the Visualization Toolkit (VTK), an application-independent scientific visualization library.
PURPOSE: Rhabdomyosarcoma (RMS) is an aggressive soft-tissue sarcoma which primarily occurs in ch... more PURPOSE: Rhabdomyosarcoma (RMS) is an aggressive soft-tissue sarcoma which primarily occurs in children and young adults. We previously reported specific genomic alterations in RMS which strongly correlated with survival; however, predicting these mutations or high-risk disease at diagnosis remains a significant challenge. In this study, we utilized convolutional neural networks (CNNs) to learn histologic features associated with driver mutations and outcome using H&E images of RMS. PATIENTS AND METHODS: Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from n=321 RMS patients enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n=136) or holdout test data. RESULTS: The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared to current molecular-clinical risk stratification. CONCLUSION: This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a
ObjectiveBiomechanical simulation requires accurate representation of the geometry of the structu... more ObjectiveBiomechanical simulation requires accurate representation of the geometry of the structures to be studied. Specifically, when simulating the interaction of implanted prosthetics with surrounding tissues, the geometric and mechanical properties of the prosthesis and the surrounding tissues need to be adequately represented. The present work describes methods for inserting test prostheses into magnetic resonance imaging (MRI) derived geometric models of the pelvic floor structures, in order to create realistic simulation models.MethodsWe modified an existing public domain image analysis software tool to allow placement and segmentation of arbitrarily shaped 3D objects into the output segmented geometry of a pelvic MRI image dataset. The tool was applied to create composite segmented geometry and 3D models of the MRI derived pelvic floor structures with the inserted prostheses in the intended anatomic locations, suitable for biomechanical simulation model creation.ResultsSegmentations of the organs in the source pelvic MRI datasets were created, showing the segmented embedded prostheses in the planned location. Three dimensional reconstructions of the segmented datasets were generated, which were viewable from multiple angles, and the ability to turn on and off all tissue and prosthesis layers was demonstrated.ConclusionsWe created a software application for inserting prostheses into segmented MRI based datasets. The output segmentations were suitable for input into a soft-tissue simulation tool suite, which generated simulation results suitable for analysis. This tool has the potential to enable patient specific, iterative surgical planning of prolapse repair strategies.Funding Source(s)None
Background Mouse models are highly effective for studying the pathophysiology of lung adenocarcin... more Background Mouse models are highly effective for studying the pathophysiology of lung adenocarcinoma and evaluating new treatment strategies. Treatment efficacy is primarily determined by the total tumor burden measured on excised tumor specimens. The measurement process is time-consuming and prone to human errors. To address this issue, we developed a novel deep learning model to segment lung tumor foci on digitally scanned hematoxylin and eosin (H&E) histology slides. Methods Digital slides of 239 mice from 9 experimental cohorts were split into training (n=137), validation (n=37), and testing cohorts (n=65). Image patches of 500×500 pixels were extracted from 5× and 10× magnifications, along with binary masks of expert annotations representing ground-truth tumor regions. Deep learning models utilizing DeepLabV3+ and UNet architectures were trained for binary segmentation of tumor foci under varying stain normalization conditions. The performance of algorithm segmentation was assessed by Dice Coefficient, and detection was evaluated by sensitivity and positive-predictive value (PPV). Results The best model on patch-based validation was DeepLabV3+ using a Resnet-50 backbone, which achieved Dice 0.890 and 0.873 on validation and testing cohort, respectively. This result corresponded to 91.3 Sensitivity and 51.0 PPV in the validation cohort and 93.7 Sensitivity and 51.4 PPV in the testing cohort. False positives could be reduced 10-fold with thresholding artificial intelligence (AI) predicted output by area, without negative impact on Dice Coefficient. Evaluation at various stain normalization strategies did not demonstrate improvement from the baseline model. Conclusions A robust AI-based algorithm for detecting and segmenting lung tumor foci in the pre-clinical mouse models was developed. The output of this algorithm is compatible with open-source software that researchers commonly use.
While many are familiar with flight simulators, there is also a growing body of ground-based simu... more While many are familiar with flight simulators, there is also a growing body of ground-based simulation training systems. The Army/DARPA sponsored SIMNET project (Nelms 1988) involved over 200 armor and aircraft simulators in a complex network, designed to teach combined arms combat skills. The follow-on Close Combat Tactical Trainer (CCTT) project will be the largest training simulator acquisition in history. In addition, realtime interactive simula tion is moving beyond military training into the potentially much larger market of commercial, entertainment and educational applications currently being called "Virtual Reality" (Furness 1988). However, no existing realtime simulation supports a truly interactive world. In particular, the terrain (soil, water and vegetation) is nearly or completely immutable in today's simula tors. In a word, the terrain is not dynamic. This Project explores the hypothesis that it is economically feasible to construct networked realtime simulators which incorporate useful simulations of dynamic terrain phenomena. The authors have evaticated the computational requirements of realtime graphical dynamic terrain simula tion with both theoretical models and prototypes, and conclude that useful levels of terrain dynamics can be incorporated in the next generation of low-cost, high-volume training simulators and virtual environ ments.
bioRxiv (Cold Spring Harbor Laboratory), Dec 6, 2023
This study presents an innovative approach for understanding the genetic underpinnings of two key... more This study presents an innovative approach for understanding the genetic underpinnings of two key phenotypes in Sorghum bicolor: maximum canopy height and maximum growth rate. Genome-Wide Association Studies (GWAS) are widely used to decipher the genetic basis of traits in organisms, but the challenge lies in selecting an appropriate statistically significant threshold for analysis. Our goal was to employ GWAS to pinpoint the genetic markers associated with the phenotypes of interest using specific permissive-filtered threshold values that allows the inclusion of broader collections of explanatory candidate genes. Then, we utilized a pattern recognition technique to prioritize a set of informative genes, which hold potential for further investigation and could find applications in Artificial Intelligence systems. Utilizing a subset of the Sorghum Bioenergy Association Panel cultivated at the Maricopa Agricultural Center in Arizona, we sought to unveil patterns between phenotypic similarity and genetic proximity among accessions in order to organize Single Nucleotide Polymorphisms (SNPs) which are likely to be associated with the phenotypic trait. Additionally, we explored the impact of this method by considering all SNPs versus focusing on SNPs classified through the GWAS pre-filter. Experimental results indicated that our approach effectively prioritizes SNPs and genes influencing the phenotype of interest. Moreover, this methodology holds promise in the feature selection from genomic data for predicting complex phenotypic traits influenced by numerous genes and environmental conditions and could pave the way for further research in this field.
Members of different species often engage in aggressive contests over resources. This series of a... more Members of different species often engage in aggressive contests over resources. This series of aggressive contests between species may result in an interspecific dominance hierarchy. Such hierarchies are of interest because they could be used to address a variety of research questions, for example, do similarly ranked species tend to avoid each other in time or space, and what will happen when such species come into contact as climates change? Here, we propose a method for creating a continental-scale hierarchy, and we make initial analyses based on this hierarchy. Leveraging the existing network of citizen scientists from Project FeederWatch, we collected the data with which to create a continent-spanning interspecific dominance hierarchy that included species that do not currently have overlapping geographic distributions. We quantified the extent of intransitivities (rock-paper-scissors relationships) in the hierarchy, as intransitivities can promote local species’ coexistence. Overall, the hierarchy was nearly linear, and largely predicted by body mass, although there were clade-specific deviations from the average mass–dominance relationship. Warblers and orioles, for instance, were more dominant than expected based on their body mass, while buntings, grosbeaks, and doves were less dominant than expected. Intransitive relationships were rare. Few interactions were reported between close relatives and ecological competitors like Mountain and Black-capped Chickadees, as such species often have only marginally overlapping geographic distributions, restricting opportunity for observation. Yet, these species’ ranks—emergent properties of the network—were often in agreement with targeted studies of dominance relationships between them
The study of aggressive interactions between species has, to date, usually been restricted to int... more The study of aggressive interactions between species has, to date, usually been restricted to interactions among small numbers of ecologically close competitors. Nothing is known about interspecific dominance hierarchies that include numerous, ecologically varied species. Such hierarchies are of interest because they could be used to address a variety of research questions, e.g. do similarly ranked species tend to avoid each other in time or Interspecific dominance hierarchy, aggression, displacement, citizen science LAY SUMMARY When it comes to fighting over food, bigger is better but woodpeckers are best. The outcome of aggressive encounters between birds frequently determines which individual gains access to contested resources like food, but until now, little was known about such encounters between individuals of different species. We partnered with citizen scientists to record interspecific behavioral interactions at bird feeders around North America, and assembled these interactions into a continental dominance hierarchy. .
Supplemental Table S1. Whole slide image tissue segmentation statistics by an expert pathologist ... more Supplemental Table S1. Whole slide image tissue segmentation statistics by an expert pathologist and probability prediction using a trained convolutional neural network. Supplemental Table S2. Clinical and molecular characteristics of FN-RMS samples used for training models for mutation prediction. Yellow boxes indicate genes included in defining the RAS pathway. Supplemental Table S3. Clinical information with COG risk stratification of FN-RMS samples used for training a prognostication predictive CNN.
A.I. classification of duplicate cores Match Mismatch Total Number of patients 109 13 122 Proport... more A.I. classification of duplicate cores Match Mismatch Total Number of patients 109 13 122 Proportion (95% C.
Supplemental Figure S6. Graphical User Interface for ssue segmenta on, MYOD1 muta on predic on, a... more Supplemental Figure S6. Graphical User Interface for ssue segmenta on, MYOD1 muta on predic on, and risk predic on models. (a) A containerized docker was developed to allow users to upload H&E images of RMS ssue and perform ssue segmenta on to classify regions of tumor. (b) Propor ons of ssue predicted as ARMS, ERMS, necrosis, and stroma are generated as a bar graph. (c) User-provided images can also be analyzed for MYOD1 muta ons using our trained MYOD1 muta on predic on model. The posi ve predic on score can be compared with our training cohort of samples that have known MYOD1 muta on status. (d) A hazard predic on score (-1 to +1) can be a ained using the trained FN-RMS risk predic on model. The hazard predic on score can be compared with A.I. predicted risk grouping for our study cohort data (n=264).
Purpose: Rhabdomyosarcoma (RMS) is an aggressive softtissue sarcoma, which primarily occurs in ch... more Purpose: Rhabdomyosarcoma (RMS) is an aggressive softtissue sarcoma, which primarily occurs in children and young adults. We previously reported specific genomic alterations in RMS, which strongly correlated with survival; however, predicting these mutations or high-risk disease at diagnosis remains a significant challenge. In this study, we utilized convolutional neural networks (CNN) to learn histologic features associated with driver mutations and outcome using hematoxylin and eosin (H&E) images of RMS. Experimental Design: Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from 321 patients with RMS enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n ¼ 136) or holdout test data. Results: The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared with current molecular-clinical risk stratification. Conclusions: This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a powerful tool for diagnostic and prognostic prediction of rhabdomyosarcoma, which will be tested in prospective COG clinical trials.
Abstract This chapter is designed to help the reader become familiar with standard methods for de... more Abstract This chapter is designed to help the reader become familiar with standard methods for describing pelvic organ geometry, suitable for biomechanical computational analysis. To perform structural engineering analysis of the pelvic floor tissues, they must first be represented geometrically. This is done via a technique called segmentation, in which individual organs and/or tissue elements are outlined, and given unique labels. The interesting tissues are usually identified on standard radiologic image stacks — usually magnetic resonance imaging (MRI) or computed tomography (CT) scans. The goal of segmentation is to identify a desired tissue on the original (grayscale) image, and trace out (or label) its boundaries. For each organ or tissue of interest, the labeling is performed, either manually or semiautomatically. Because pelvic floor structures exist in a three-dimensional (3-D) space, it is necessary to represent the tissues as 3-D structures. This 3-D representation is facilitated by the method of acquiring MRI and CT image data. Both CT and MRI will produce grayscale source image data sets. Each data set consists of a “stack” of 2-D images, covering the region of the body that was scanned. The extent of the scanned region is called the field of view, and the grayscale images are sometimes referred to as the source images. This chapter will discuss the geometric representation of pelvic floor MRI grayscale data, as well as the segmentation of this data to produce labelmaps suitable for computational analysis and 3-D reconstruction of the organs and tissues of interest. Manual, semiautomatic, and automatic segmentation methods will be briefly introduced, in preparation for a more detailed discussion of automated segmentation in a later chapter. The chapter will conclude with a review of the trade-offs between the segmentation methods, and discussion of the issues with repeatability, reliability, and fidelity of these methods. The chapter focuses on MRI source data, with the understanding that the general principles will apply to CT sourced images as well.
In this paper, we present a molecular modeling approach based on recursive ot~ect-oriented class ... more In this paper, we present a molecular modeling approach based on recursive ot~ect-oriented class instances. Our approach allows ease of use by the scientist and includes a heuristic for evaluating molecular interaction and rendering at multiple levels of detail. Our implementation in C++ employs abstractions to encapsulate the implementation and details of the computations. Visualization is accomplished with OpenGL or the Visualization Toolkit (VTK), an application-independent scientific visualization library.
PURPOSE: Rhabdomyosarcoma (RMS) is an aggressive soft-tissue sarcoma which primarily occurs in ch... more PURPOSE: Rhabdomyosarcoma (RMS) is an aggressive soft-tissue sarcoma which primarily occurs in children and young adults. We previously reported specific genomic alterations in RMS which strongly correlated with survival; however, predicting these mutations or high-risk disease at diagnosis remains a significant challenge. In this study, we utilized convolutional neural networks (CNNs) to learn histologic features associated with driver mutations and outcome using H&E images of RMS. PATIENTS AND METHODS: Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from n=321 RMS patients enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n=136) or holdout test data. RESULTS: The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared to current molecular-clinical risk stratification. CONCLUSION: This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a
ObjectiveBiomechanical simulation requires accurate representation of the geometry of the structu... more ObjectiveBiomechanical simulation requires accurate representation of the geometry of the structures to be studied. Specifically, when simulating the interaction of implanted prosthetics with surrounding tissues, the geometric and mechanical properties of the prosthesis and the surrounding tissues need to be adequately represented. The present work describes methods for inserting test prostheses into magnetic resonance imaging (MRI) derived geometric models of the pelvic floor structures, in order to create realistic simulation models.MethodsWe modified an existing public domain image analysis software tool to allow placement and segmentation of arbitrarily shaped 3D objects into the output segmented geometry of a pelvic MRI image dataset. The tool was applied to create composite segmented geometry and 3D models of the MRI derived pelvic floor structures with the inserted prostheses in the intended anatomic locations, suitable for biomechanical simulation model creation.ResultsSegmentations of the organs in the source pelvic MRI datasets were created, showing the segmented embedded prostheses in the planned location. Three dimensional reconstructions of the segmented datasets were generated, which were viewable from multiple angles, and the ability to turn on and off all tissue and prosthesis layers was demonstrated.ConclusionsWe created a software application for inserting prostheses into segmented MRI based datasets. The output segmentations were suitable for input into a soft-tissue simulation tool suite, which generated simulation results suitable for analysis. This tool has the potential to enable patient specific, iterative surgical planning of prolapse repair strategies.Funding Source(s)None
Background Mouse models are highly effective for studying the pathophysiology of lung adenocarcin... more Background Mouse models are highly effective for studying the pathophysiology of lung adenocarcinoma and evaluating new treatment strategies. Treatment efficacy is primarily determined by the total tumor burden measured on excised tumor specimens. The measurement process is time-consuming and prone to human errors. To address this issue, we developed a novel deep learning model to segment lung tumor foci on digitally scanned hematoxylin and eosin (H&E) histology slides. Methods Digital slides of 239 mice from 9 experimental cohorts were split into training (n=137), validation (n=37), and testing cohorts (n=65). Image patches of 500×500 pixels were extracted from 5× and 10× magnifications, along with binary masks of expert annotations representing ground-truth tumor regions. Deep learning models utilizing DeepLabV3+ and UNet architectures were trained for binary segmentation of tumor foci under varying stain normalization conditions. The performance of algorithm segmentation was assessed by Dice Coefficient, and detection was evaluated by sensitivity and positive-predictive value (PPV). Results The best model on patch-based validation was DeepLabV3+ using a Resnet-50 backbone, which achieved Dice 0.890 and 0.873 on validation and testing cohort, respectively. This result corresponded to 91.3 Sensitivity and 51.0 PPV in the validation cohort and 93.7 Sensitivity and 51.4 PPV in the testing cohort. False positives could be reduced 10-fold with thresholding artificial intelligence (AI) predicted output by area, without negative impact on Dice Coefficient. Evaluation at various stain normalization strategies did not demonstrate improvement from the baseline model. Conclusions A robust AI-based algorithm for detecting and segmenting lung tumor foci in the pre-clinical mouse models was developed. The output of this algorithm is compatible with open-source software that researchers commonly use.
While many are familiar with flight simulators, there is also a growing body of ground-based simu... more While many are familiar with flight simulators, there is also a growing body of ground-based simulation training systems. The Army/DARPA sponsored SIMNET project (Nelms 1988) involved over 200 armor and aircraft simulators in a complex network, designed to teach combined arms combat skills. The follow-on Close Combat Tactical Trainer (CCTT) project will be the largest training simulator acquisition in history. In addition, realtime interactive simula tion is moving beyond military training into the potentially much larger market of commercial, entertainment and educational applications currently being called "Virtual Reality" (Furness 1988). However, no existing realtime simulation supports a truly interactive world. In particular, the terrain (soil, water and vegetation) is nearly or completely immutable in today's simula tors. In a word, the terrain is not dynamic. This Project explores the hypothesis that it is economically feasible to construct networked realtime simulators which incorporate useful simulations of dynamic terrain phenomena. The authors have evaticated the computational requirements of realtime graphical dynamic terrain simula tion with both theoretical models and prototypes, and conclude that useful levels of terrain dynamics can be incorporated in the next generation of low-cost, high-volume training simulators and virtual environ ments.
Uploads
Papers by Curtis Lisle