Florent Marie :"An innovative platform for the segmentation of medical images by Case-Based Reasoning (CBR) and Deep Learning methods"

Thusday December 19th - 2.00 PM

Amp

Travaux de thèse de Florent Marie : "An innovative platform for the segmentation of medical images by Case-Based Reasoning (CBR) and Deep Learning methods"

Abstract :

Nephroblastoma (or Wilms' tumor) aects the kidney and is the most common childhood abdominal cancer tumor. During therapy, it is recommended to preserve kidney function as well as possible by preserving the pathological kidney under certain conditions. The information from the imagery is important to validate them, but partly involves segmentation of the scan images. As this task is very time-consuming, it is common for this information not to be fully exploited. Articial Intelligence (AI) represents a promising way to automate segmentations but generally requires a large amount of learning data. Among other things, Case-Based Reasoning (CBR) is a knowledge-based AI approach that integrates business knowledge and an adaptation of an existing solution to optimize the resolution of a new problem. Convolutional Neutral Networks (CNN) are a purely experiential approach. We propose a CBR system coupled with a region growing algorithm in order to segment kidneys
deformed by nephroblastoma. In parallel, a training method for CNN, called OV 2ASSION, is also proposed for tumor segmentation. Both approaches aim to addressing a lack of learning data (few segmentations of nephroblastoma and pathological kidneys are available). The evaluations show good performance with the calculation of segmentations similar to those made by radiologists and surgeons.

Jury composition :