Home // International Journal On Advances in Life Sciences, volume 17, numbers 1 and 2, 2025 // View article
Authors:
Anil Gavade
Rajendra Nerli
Shridhar Ghagane
Les Sztandera
Keywords: prostate cancer; gleason grading; computational pathology; attention mechanisms; whole-slide imaging; clinical decision support.
Abstract:
Prostate cancer (PCa) remains a leading cause of cancer-related mortality in urologic oncology, with the prostate gland being both the primary site of tumorigenesis and a key determinant of disease progression. Histopathological evaluation continues to be the gold standard for diagnosis, based on systematic biopsy protocols and Gleason grading (GG), which classifies acinar architectural patterns. Modern diagnostic workflows incorporate multiparametric MRI (mpMRI) and PI-RADS scoring for targeted lesion sampling, while advanced techniques such as whole-mount section analysis of radical prostatectomy specimens enable comprehensive tumor mapping. Immunohistochemical markers further aid in resolving diagnostic ambiguities, informing risk stratification and treatment decisions through assessment of tumor volume, perineural invasion, and margin status. However, GG is limited by inter-observer variability, labor-intensive interpretation, and lack of expert pathologists in resource-constrained settings. To address these challenges, we introduce LightGleason, a lightweight and interpretable deep learning framework that converts subjective Gleason grading into an objective, automated process. The model architecture combines a MobileNetV2 backbone for efficient local feature extraction with a gated multi-head self-attention (MHSA) mechanism that emphasizes diagnostically critical tissue regions, enabling improved differentiation between closely related Gleason patterns such as grade groups 3 and 4, while reducing computational overhead by 38%. Trained and validated on the SICAPv2 dataset, which includes 2,186 expert-annotated whole-slide images from three institutions, LightGleason achieves 96.8% classification accuracy—outperforming conventional models such as ResNet50, InceptionV3, and Xception by 3–7%. Ablation studies highlight the effectiveness of MHSA in enhancing F1-scores for high-grade tumors and improving resilience to histological artifacts. Simulated clinical trials demonstrate a 70% reduction in diagnostic time, making the system suitable for integration into clinical workflows. LightGleason offers an efficient, explainable, and clinically deployable solution that enhances diagnostic accuracy, speeds up clinical decision-making, and supports the standardization of prostate cancer grading across diverse healthcare environments.
Pages: 41 to 55
Copyright: Copyright (c) to authors, 2025. Used with permission.
Publication date: June 30, 2025
Published in: journal
ISSN: 1942-2660