2021/06/25 by Gabriel Garcı́a, Anna Esteve, García, Gabriel +7
Medicine · #Bladder and Urothelial Cancer Treatments #Colorectal Cancer Screening and Detection #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Radiomics and Machine Learning in Medical Imaging #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2106.13559
openalex publication_date 2021/06/25 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Recently, bladder cancer has been significantly increased in terms of\nincidence and mortality. Currently, two subtypes are known based on tumour\ngrowth: non-muscle invasive (NMIBC) and muscle-invasive bladder cancer (MIBC).\nIn this work, we focus on the MIBC subtype because it is of the worst prognosis\nand can spread to adjacent organs. We present a self-learning framework to\ngrade bladder cancer from histological images stained via immunohistochemical\ntechniques. Specifically, we propose a novel Deep Convolutional Embedded\nAttention Clustering (DCEAC) which allows classifying histological patches into\ndifferent severity levels of the disease, according to the patterns established\nin the literature. The proposed DCEAC model follows a two-step fully\nunsupervised learning methodology to discern between non-tumour, mild and\ninfiltrative patterns from high-resolution samples of 512x512 pixels. Our\nsystem outperforms previous clustering-based methods by including a\nconvolutional attention module, which allows refining the features of the\nlatent space before the classification stage. The proposed network exceeds\nstate-of-the-art approaches by 2-3% across different metrics, achieving a final\naverage accuracy of 0.9034 in a multi-class scenario. Furthermore, the reported\nclass activation maps evidence that our model is able to learn by itself the\nsame patterns that clinicians consider relevant, without incurring prior\nannotation steps. This fact supposes a breakthrough in muscle-invasive bladder\ncancer grading which bridges the gap with respect to train the model on\nlabelled data.\n