Benchmarking Image Transformers for Prostate Cancer Detection from Ultrasound Data
Authors:
Mohamed Harmanani,
Paul F. R. Wilson,
Fahimeh Fooladgar,
Amoon Jamzad,
Mahdi Gilany,
Minh Nguyen Nhat To,
Brian Wodlinger,
Purang Abolmaesumi,
Parvin Mousavi
Abstract:
PURPOSE: Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional networks (CNNs) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches h…
▽ More
PURPOSE: Deep learning methods for classifying prostate cancer (PCa) in ultrasound images typically employ convolutional networks (CNNs) to detect cancer in small regions of interest (ROI) along a needle trace region. However, this approach suffers from weak labelling, since the ground-truth histopathology labels do not describe the properties of individual ROIs. Recently, multi-scale approaches have sought to mitigate this issue by combining the context awareness of transformers with a CNN feature extractor to detect cancer from multiple ROIs using multiple-instance learning (MIL). In this work, we present a detailed study of several image transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and transformers for ultrasound-based prostate cancer classification. We also design a novel multi-objective learning strategy that combines both ROI and core predictions to further mitigate label noise. METHODS: We evaluate 3 image transformers on ROI-scale cancer classification, then use the strongest model to tune a multi-scale classifier with MIL. We train our MIL models using our novel multi-objective learning strategy and compare our results to existing baselines. RESULTS: We find that for both ROI-scale and multi-scale PCa detection, image transformer backbones lag behind their CNN counterparts. This deficit in performance is even more noticeable for larger models. When using multi-objective learning, we can improve performance of MIL, with a 77.9% AUROC, a sensitivity of 75.9%, and a specificity of 66.3%. CONCLUSION: Convolutional networks are better suited for modelling sparse datasets of prostate ultrasounds, producing more robust features than transformers in PCa detection. Multi-scale methods remain the best architecture for this task, with multi-objective learning presenting an effective way to improve performance.
△ Less
Submitted 26 March, 2024;
originally announced March 2024.
Modelling the Spread of COVID-19 in Indoor Spaces using Automated Probabilistic Planning
Authors:
Mohamed Harmanani
Abstract:
The coronavirus disease 2019 (COVID-19) pandemic has been ongoing for around 3 years, and has infected over 750 million people and caused over 6 million deaths worldwide at the time of writing. Throughout the pandemic, several strategies for controlling the spread of the disease have been debated by healthcare professionals, government authorities, and international bodies. To anticipate the poten…
▽ More
The coronavirus disease 2019 (COVID-19) pandemic has been ongoing for around 3 years, and has infected over 750 million people and caused over 6 million deaths worldwide at the time of writing. Throughout the pandemic, several strategies for controlling the spread of the disease have been debated by healthcare professionals, government authorities, and international bodies. To anticipate the potential impact of the disease, and to simulate the effectiveness of different mitigation strategies, a robust model of disease spread is needed. In this work, we explore a novel approach based on probabilistic planning and dynamic graph analysis to model the spread of COVID-19 in indoor spaces. We endow the planner with means to control the spread of the disease through non-pharmaceutical interventions (NPIs) such as mandating masks and vaccines, and we compare the impact of crowds and capacity limits on the spread of COVID-19 in these settings. We demonstrate that the use of probabilistic planning is effective in predicting the amount of infections that are likely to occur in shared spaces, and that automated planners have the potential to design competent interventions to limit the spread of the disease. Our code is fully open-source and is available at: https://github.com/mharmanani/prob-planning-covid19 .
△ Less
Submitted 16 August, 2023;
originally announced August 2023.
