Predição de casos de comprometimento cognitivo leve estável e progressivo usando atributos de imagens de ressonância magnética para auxílio no diagnóstico da doença de Alzheimer

Abstract

Alzheimer’s Disease (AD) is a progressive and irreversible brain disorder that affects cognitive functions, accounting for 60 to 80% of dementia cases worldwide. With approximately 50 million people affected, AD is considered a major public health issue. Although there is no cure, treatments can slow its progression, especially when applied in the early stages. In this context, the prediction of Mild Cognitive Impairment (MCI) cases in its stable and progressive forms is crucial for more effective interventions. This study aims to develop a predictive model based on the Random Forest algorithm using features extracted from magnetic resonance imaging (MRI), such as cortical thickness, volume, and texture from different brain regions, to aid in the early diagnosis of AD. The ComBat harmonization method was applied to ensure model reproducibility. Model performance was assessed using metrics such as accuracy, balanced accuracy, sensitivity, specificity, Area Under the Curve (AUC) and F1-score, allowing for a detailed analysis of their effectiveness in predicting AD progression based on MRI scans. The best F1-score performance was achieved using feature selection based on Random Forest importance, with the 75th percentile as the cutoff. A total of 312 features across 66 Regions of Interest (ROIs) were selected. The results were: F1-score 84.31±8.73%, accuracy 84.31±8.70%, balanced accuracy 84.31±8.64%, sensitivity 84.00±9.67%, specificity 84.62±13.01%, and AUC 0.85±0.07. The results demonstrate that the proposed approach combines competitive performance and interpretability, offering a promising and clinically relevant tool for predicting the conversion from MCI to AD based on MRI.