Explaining the Image Content via Machine Learning Techniques: Extending the Experience on Semantic Labeling from Remote Sensing to Medical Imaging

Abstract

This presentation explores the application of machine learning (ML) techniques to interpret content from Earth observation (EO) and medical imaging data. Our objective was to leverage Support Vector Machine (SVM) algorithms to discover semantic relations in diverse image datasets. We collated EO images utilizing multispectral and radar sensors from four urban areas and procured medical images via camera, microscope, and computed tomography (CT). Our methodology underwent rigorous testing by domain experts, with classification results validated against reference datasets and expert evaluations, achieving 95% accuracy for satellite images and 85% for medical images. We applied our approach to correlate EO images detailing environmental data impacting quality of life with medical images indicating disease phenotypes, aiming to enhance epidemiological studies' precision.