FAE

FAE facilitates development and evaluation of supervised machine learning models for radiomics and medical image analysis.

Key Features:

• Implementation: Implemented in Python and leveraging NumPy, pandas, and scikit-learn for data handling, analysis, and modeling.
• Feature Extraction and Preprocessing: Extraction of image features and preprocessing of the resulting feature matrix, including preparatory steps for analysis.
• Normalization: Standardization of feature values to ensure consistency across datasets.
• Automated Model Development: Automatic development and exploration of multiple supervised machine learning models with configurable combinations of feature selectors and classifiers.
• Feature Selection Support: Support for feature selection techniques such as analysis of variance (ANOVA).
• Classification Support: Support for classifiers including linear discriminant analysis (LDA).
• Model Evaluation and Metrics: Evaluation of models using clinical statistics and performance metrics including area under the receiver operating characteristic curve (AUC).
• Model Comparison and Visualization: Comparison of models via visualization of key performance metrics such as AUC.

Scientific Applications:

• Radiomics outcome mapping: Mapping image-derived features to clinical outcomes in radiomics studies.
• Supervised learning for medical imaging: Application of supervised machine learning workflows to medical studies that use imaging data.
• Prostate cancer classification (PROSTATEx): Classification of clinically significant prostate cancer (CS PCa) versus non-CS PCa on the PROSTATEx dataset using ANOVA for feature selection and LDA for classification, yielding AUCs of 0.838 (training), 0.814 (validation), and 0.824 (test).

Methodology:

Normalization, feature selection using analysis of variance (ANOVA), and classification using linear discriminant analysis (LDA).