Prediction of Patient Survival Through Serial and Multimodal Analysis of Medical Exams with AI

• Grant number: 25/14527-8
• Support Opportunities: Scholarships in Brazil - Doctorate (Direct)
• Start date: September 01, 2025
• End date: August 31, 2028
• Field of knowledge: Health Sciences - Medicine - Medical Radiology
• Principal Investigator: Giovanni Guido Cerri
• Grantee: Lucas Vinicius Buchelt Souza
• Host Institution: Faculdade de Medicina (FM). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 21/11905-0 - Center of Science, Technology and Development for innovation in Medicine and Health: inLab.iNova, AP.CCD

Abstract

This PhD thesis proposes the development of a serial and multimodal analysis model with Artificial Intelligence (AI) for predicting patient survival. The aim is to integrate structured and unstructured information using neural network models. The motivation for this thesis stems from the recent technological increase and the limitations of traditional survival prediction models, since they analyze data in a punctual way and do not decode the patient's temporal evolution. There are also few multimodal approaches for this purpose. In this thesis I propose the use of databases such as MIMIC-IV, with anonymized data from more than 60,000 patients, with an intermediate pre-processing and data fusion stage. As the main data analysis architecture, I propose Long Short Term Memory, together with attention mechanisms for serial analysis. A secondary objective is the concept of Explainable Artificial Intelligence, in which the results from the prediction agorithm can be interpreted in a clinical context. The originality of the thesis project lies in the combination of temporal analysis, multimodality and survival prediction approaches.