Využití neuronových sítí pro detekci a hodnocení extrakapsulárního šíření karcinomu prostaty s využitím interpretovatelných metod umělé inteligence

Abstract

This master’s thesis explores the application of neural networks for the detection and assessment of extraprostatic extension using explainable artificial intelligence methods. The theoretical section addresses the clinical challenges extraprostatic extension, medical image processing including seg mentation and the architectures of neural networks commonly used for these tasks. It also provides an overview of XAI techniques for medical imaging. The practical part of the thesis involves the development of a segmentation model trained on a publicly available dataset. The training and validation process resulted in five models, which were subsequently utilized for ensemble learning. A clinical dataset provided by the University Hospital Ostrava was then processed through these models, followed by the application of the Grad-CAM method to visualize the decision-making pro cess.The results demonstrate that the model is capable of segmenting the prostate on unseen clinical data. Moreover, the generated heatmaps confirmed that the model’s focus is directed toward the prostatic capsule, which is the area for extraprostatic spread.