Home // AIHealth 2025, The Second International Conference on AI-Health // View article
Authors:
Anil Gavade
Rajendra Nerli
Shridhar Ghagane
Les Sztandera
Keywords: Cancer diagnosis; Dimensionality reduction; Explainable AI; Feature extraction; Gleason grade classification.
Abstract:
Accurate grading of Prostate Cancer (PCa) is vital for effective treatment planning and prognosis. This study introduces an advanced framework for Gleason Grade (GG) classification, addressing challenges in accuracy, computational efficiency, and interpretability. Utilizing the SICAPv2 dataset, which contains annotated prostate biopsy Whole Slide Images (WSIs) graded from GG0 to GG5, the framework integrates cutting-edge machine learning and deep learning techniques. Feature extraction is performed using a custom-designed Variational Autoencoder (VAE) with a VGG16 backbone, chosen for its computational efficiency, while dimensionality reduction with Principal Component Analysis (PCA) optimally selects 50 features for classification. The classification pipeline combines machine learning models, including Support Vector Machines (SVM), logistic regression, and random forests, with custom Deep Neural Networks (DNNs). SVM with an Radial Basis Function (RBF) kernel achieved an accuracy of 84% following hyperparameter tuning, while a custom five-layer dense neural network incorporating dropout and batch normalization demonstrated superior performance with an accuracy of 94.6%. Explainable AI (XAI) techniques, such as SHapley Additive exPlanations (SHAP), gradient-weighted class activation mapping (Grad-CAM), and Local Interpretable Model-Agnostic Explanations (LIME), enhance model interpretability by providing insights into feature importance and aligning predictions with clinical expertise. This framework delivers a robust, scalable, and interpretable solution for automated GG classification, bridging the gap between advanced AI techniques and clinical application.
Pages: 1 to 6
Copyright: Copyright (c) IARIA, 2025
Publication date: March 9, 2025
Published in: conference
ISBN: 978-1-68558-247-0
Location: Lisbon, Portugal
Dates: from March 9, 2025 to March 13, 2025