-
CAManim: Animating end-to-end network activation maps
Authors:
Emily Kaczmarek,
Olivier X. Miguel,
Alexa C. Bowie,
Robin Ducharme,
Alysha L. J. Dingwall-Harvey,
Steven Hawken,
Christine M. Armour,
Mark C. Walker,
Kevin Dick
Abstract:
Deep neural networks have been widely adopted in numerous domains due to their high performance and accessibility to developers and application-specific end-users. Fundamental to image-based applications is the development of Convolutional Neural Networks (CNNs), which possess the ability to automatically extract features from data. However, comprehending these complex models and their learned rep…
▽ More
Deep neural networks have been widely adopted in numerous domains due to their high performance and accessibility to developers and application-specific end-users. Fundamental to image-based applications is the development of Convolutional Neural Networks (CNNs), which possess the ability to automatically extract features from data. However, comprehending these complex models and their learned representations, which typically comprise millions of parameters and numerous layers, remains a challenge for both developers and end-users. This challenge arises due to the absence of interpretable and transparent tools to make sense of black-box models. There exists a growing body of Explainable Artificial Intelligence (XAI) literature, including a collection of methods denoted Class Activation Maps (CAMs), that seek to demystify what representations the model learns from the data, how it informs a given prediction, and why it, at times, performs poorly in certain tasks. We propose a novel XAI visualization method denoted CAManim that seeks to simultaneously broaden and focus end-user understanding of CNN predictions by animating the CAM-based network activation maps through all layers, effectively depicting from end-to-end how a model progressively arrives at the final layer activation. Herein, we demonstrate that CAManim works with any CAM-based method and various CNN architectures. Beyond qualitative model assessments, we additionally propose a novel quantitative assessment that expands upon the Remove and Debias (ROAD) metric, pairing the qualitative end-to-end network visual explanations assessment with our novel quantitative "yellow brick ROAD" assessment (ybROAD). This builds upon prior research to address the increasing demand for interpretable, robust, and transparent model assessment methodology, ultimately improving an end-user's trust in a given model's predictions.
△ Less
Submitted 18 December, 2023;
originally announced December 2023.
-
MetaCAM: Ensemble-Based Class Activation Map
Authors:
Emily Kaczmarek,
Olivier X. Miguel,
Alexa C. Bowie,
Robin Ducharme,
Alysha L. J. Dingwall-Harvey,
Steven Hawken,
Christine M. Armour,
Mark C. Walker,
Kevin Dick
Abstract:
The need for clear, trustworthy explanations of deep learning model predictions is essential for high-criticality fields, such as medicine and biometric identification. Class Activation Maps (CAMs) are an increasingly popular category of visual explanation methods for Convolutional Neural Networks (CNNs). However, the performance of individual CAMs depends largely on experimental parameters such a…
▽ More
The need for clear, trustworthy explanations of deep learning model predictions is essential for high-criticality fields, such as medicine and biometric identification. Class Activation Maps (CAMs) are an increasingly popular category of visual explanation methods for Convolutional Neural Networks (CNNs). However, the performance of individual CAMs depends largely on experimental parameters such as the selected image, target class, and model. Here, we propose MetaCAM, an ensemble-based method for combining multiple existing CAM methods based on the consensus of the top-k% most highly activated pixels across component CAMs. We perform experiments to quantifiably determine the optimal combination of 11 CAMs for a given MetaCAM experiment. A new method denoted Cumulative Residual Effect (CRE) is proposed to summarize large-scale ensemble-based experiments. We also present adaptive thresholding and demonstrate how it can be applied to individual CAMs to improve their performance, measured using pixel perturbation method Remove and Debias (ROAD). Lastly, we show that MetaCAM outperforms existing CAMs and refines the most salient regions of images used for model predictions. In a specific example, MetaCAM improved ROAD performance to 0.393 compared to 11 individual CAMs with ranges from -0.101-0.172, demonstrating the importance of combining CAMs through an ensembling method and adaptive thresholding.
△ Less
Submitted 31 July, 2023;
originally announced July 2023.
-
Data-driven generation of 4D velocity profiles in the aneurysmal ascending aorta
Authors:
Simone Saitta,
Ludovica Maga,
Chloe Armour,
Emiliano Votta,
Declan P. O'Regan,
M. Yousuf Salmasi,
Thanos Athanasiou,
Jonathan W. Weinsaft,
Xiao Yun Xu,
Selene Pirola,
Alberto Redaelli
Abstract:
Numerical simulations of blood flow are a valuable tool to investigate the pathophysiology of ascending thoracic aortic aneurysms (ATAA). To accurately reproduce hemodynamics, computational fluid dynamics (CFD) models must employ realistic inflow boundary conditions (BCs). However, the limited availability of in vivo velocity measurements still makes researchers resort to idealized BCs. In this st…
▽ More
Numerical simulations of blood flow are a valuable tool to investigate the pathophysiology of ascending thoracic aortic aneurysms (ATAA). To accurately reproduce hemodynamics, computational fluid dynamics (CFD) models must employ realistic inflow boundary conditions (BCs). However, the limited availability of in vivo velocity measurements still makes researchers resort to idealized BCs. In this study we generated and thoroughly characterized a large dataset of synthetic 4D aortic velocity profiles suitable to be used as BCs for CFD simulations. 4D flow MRI scans of 30 subjects with ATAA were processed to extract cross-sectional planes along the ascending aorta, ensuring spatial alignment among all planes and interpolating all velocity fields to a reference configuration. Velocity profiles of the clinical cohort were extensively characterized by computing flow morphology descriptors of both spatial and temporal features. By exploiting principal component analysis (PCA), a statistical shape model (SSM) of 4D aortic velocity profiles was built and a dataset of 437 synthetic cases with realistic properties was generated. Comparison between clinical and synthetic datasets showed that the synthetic data presented similar characteristics as the clinical population in terms of key morphological parameters. The average velocity profile qualitatively resembled a parabolic-shaped profile, but was quantitatively characterized by more complex flow patterns which an idealized profile would not replicate. Statistically significant correlations were found between PCA principal modes of variation and flow descriptors. We built a data-driven generative model of 4D aortic velocity profiles, suitable to be used in computational studies of blood flow. The proposed software system also allows to map any of the generated velocity profiles to the inlet plane of any virtual subject given its coordinate set.
△ Less
Submitted 1 November, 2022;
originally announced November 2022.
