Classification of Non-alcoholic Fatty Liver Disease in Thermal Images of the Liver Using a Siamese Neural Network

• Maxwell Pires Silva UFMA
• Aristófanes Corrêa Silva UFMA
• Anselmo Cardoso de Paiva UFMA

Resumo

Non-alcoholic fatty liver disease (NAFLD) is a prevalent and severe condition that requires effective and non-invasive diagnostic methods. This work presents an innovative approach using Siamese neural networks to classify thermal images of the liver in order to identify the presence of NAFLD. The research is motivated by the need to use artificial intelligence to analyze thermal images, especially when the number of images is limited, making it difficult for ordinary neural networks to learn, a difficulty that the Siamese network already faces with ease. The proposed method involves three stages: extraction of the region of interest (ROI), image pre-processing and classification using the Siamese Neural Network, which compares pairs of images to determine their similarity and, consequently, the presence or absence of NAFLD. The preliminary results, with an accuracy of 71%, precision of 57% and recall of 96%, indicate that this approach could offer a promising tool for the non-invasive diagnosis of NAFLD, contributing a promising method to the field of machine learning applied to medicine.