A Reproducible Deep-Learning-Based Computer-Aided Diagnosis Tool for Frontotemporal Dementia Using MONAI and Clinica...

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging Analysis

Background:

• Clinical implementation of AI-based Computer-Aided Diagnosis (CAD) tools is hindered by unstandardized development practices.
• A lack of reproducible workflows for 3D MRI data hinders AI tool development in clinical settings.

Purpose of the Study:

• To develop an easily reproducible and reliable CAD tool for 3D MRI data using standardized frameworks.
• To train a Deep Learning (DL) algorithm for frontotemporal dementia (FTD) detection and ensure reproducibility.
• To apply Explainable AI (XAI) methods to understand and improve the DL model's behavior.

Main Methods:

• Utilized the Clinica and MONAI frameworks for standardized medical imaging practices.
• Trained a DL algorithm on the NIFD database for FTD detection from 3D MRI data.
• Applied attention maps from XAI to identify key brain regions influencing the DL model's decisions.

Main Results:

• The DL model achieved 0.80 accuracy, 1 sensitivity, 0.6 specificity, 0.83 F1-score, and 0.86 AUC for FTD classification.
• Explainable AI highlighted that the model's decisions were driven by hallmark brain areas associated with FTD.
• The standardized methodology provides a benchmark for FTD classification and aids in understanding model behavior.

Conclusions:

• The proposed standardized methodology facilitates reproducible development of AI-based CAD tools for neuroimaging.
• Standardization and explainability are crucial for regulatory approval and accelerating AI adoption in healthcare.
• This approach can improve the efficacy and safety evaluation of CAD tools, speeding up their integration into clinical practice.