Radiomics evaluation for the early detection of Alzheimer's dementia using T1-weighted MRI

Area of Science:

• Neuroimaging
• Machine Learning
• Radiomics

Background:

• Alzheimer's disease (AD) is a growing global health concern, necessitating early diagnosis to manage progression and enhance patient quality of life.
• Current diagnostic methods require improvement for timely detection of subtle neuroanatomical changes characteristic of early-stage AD.

Purpose of the Study:

• To detect early-stage Alzheimer's disease (AD) by identifying subtle neuroanatomical changes using Radiomics features from T1 MRI scans.
• To evaluate the efficacy of machine learning models in diagnosing AD based on extracted Radiomic features.

Main Methods:

• Utilized the AssemblyNet segmentation model on 416 anonymized T1 MRI scans to analyze brain changes.
• Extracted 1130 Radiomic features per segmentation label, totaling over 31 million features across 132 labels per patient.
• Trained and hyperparameter-tuned four machine learning models (Gradient Booster, Random Forest, Support Vector Classifier, XGBoost) using a 70/20/10 train-validation-test split, evaluated by accuracy.

Main Results:

• Achieved accuracies ranging from 0.71 to 0.86 across the four evaluated models.
• The XGBoost model demonstrated the highest accuracy of 0.86 in segmenting the left inferior lateral ventricle, indicating strong performance in early AD detection.
• Successfully segmented 208 T1-weighted MRI scans, yielding a comprehensive feature set for analysis.

Conclusions:

• Segmentation of T1-weighted MRI scans combined with Radiomics and the XGBoost model shows promising accuracy for early AD detection.
• Radiomics offers a potential pathway to significantly enhance diagnostic accuracy for AD, facilitating earlier interventions and improved patient outcomes.
• Future research should focus on expanding datasets and refining methodologies to broaden the applicability of Radiomics in AD diagnosis.