BMC medical imaging · 2025
Leveraging commonality across multiple tissue slices for enhanced whole slide image classification using graph convolutional networks.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Noree Sakonporn, Quinones Robles Willmer Rafell, Ko Young Sin, Yi Mun Yong
Аннотация
BACKGROUND: Accurate classification of histopathological whole slide images (WSIs) is essential for cancer diagnosis and treatment planning. Conventional WSI creation involves slicing a biopsy tissue into multiple slices, placing them on a single glass slide, and digitally scanning them. While deep learning approaches have shown promise in WSI analysis, they mostly overlook potential common patterns across different slices of the original tissue.
METHODS: We propose a novel technique that leverages inter-slice commonality to enhance classification performance. Our method constructs graphs for each tissue slice, extracts relevant features, and connects these graphs based on spatial relationships and feature similarities, creating a comprehensive representation of the entire tissue sample, which is then used for WSI classification using graph convolutional networks.
RESULTS: We validated our approach using stomach and colorectal WSI datasets. The results demonstrate that having the information of commonalities across slices significantly improves graph-based WSI classification models. Notably, our method outperforms existing multiple instance learning approaches in terms of both accuracy (from 87.9% to 91.5% for the stomach dataset and from 88.3% to 91.2% for the colorectal dataset), and AUROC (from 96.8% to 98.8% for the stomach dataset and from 97.3% to 98.2% for the colorectal dataset).
CONCLUSION: By efficiently establishing information across slices, our approach offers a more accurate and efficient method for WSI classification, with promising implications for clinical applications. The source code is available at https://github.com/Juckjick/commonality_graph .
CLINICAL TRIAL NUMBER: Not applicable.
METHODS: We propose a novel technique that leverages inter-slice commonality to enhance classification performance. Our method constructs graphs for each tissue slice, extracts relevant features, and connects these graphs based on spatial relationships and feature similarities, creating a comprehensive representation of the entire tissue sample, which is then used for WSI classification using graph convolutional networks.
RESULTS: We validated our approach using stomach and colorectal WSI datasets. The results demonstrate that having the information of commonalities across slices significantly improves graph-based WSI classification models. Notably, our method outperforms existing multiple instance learning approaches in terms of both accuracy (from 87.9% to 91.5% for the stomach dataset and from 88.3% to 91.2% for the colorectal dataset), and AUROC (from 96.8% to 98.8% for the stomach dataset and from 97.3% to 98.2% for the colorectal dataset).
CONCLUSION: By efficiently establishing information across slices, our approach offers a more accurate and efficient method for WSI classification, with promising implications for clinical applications. The source code is available at https://github.com/Juckjick/commonality_graph .
CLINICAL TRIAL NUMBER: Not applicable.