FeAture Explorer (FAE): A tool for developing and comparing radiomics models.

Area of Science:

• Radiomics and Medical Imaging Analysis
• Computational Biology and Bioinformatics
• Machine Learning in Healthcare

Background:

• Radiomics research typically requires developing supervised machine learning models to correlate image features with clinical outcomes.
• Constructing optimal machine learning pipelines involves multiple steps like normalization, feature selection, and classification, which can be complex and time-consuming.
• There is a need for streamlined tools to facilitate the development and evaluation of radiomics models.

Purpose of the Study:

• To introduce FeAture Explorer (FAE), an open-source software package designed to simplify and automate the development and evaluation of radiomics machine learning models.
• To provide a user-friendly graphical interface for researchers to build, compare, and visualize various machine learning pipelines.
• To demonstrate the effectiveness of FAE using a prostate cancer dataset for classifying clinically significant prostate cancer.

Main Methods:

• FAE was developed using Python, incorporating libraries such as NumPy, pandas, and scikit-learn.
• The software facilitates automatic extraction of image features, preprocessing of feature matrices, model development, and evaluation using standard clinical statistics.
• A candidate model for classifying clinically significant prostate cancer (CS PCa) and non-CS PCa was developed and evaluated on the PROSTATEx dataset, exploring various feature selectors and classifiers.

Main Results:

• FAE enabled efficient exploration of different feature selection and classification combinations for the prostate cancer dataset.
• The software facilitated visual comparison of model performance, specifically the area under the receiver operating characteristic curve (AUC), on validation and independent test datasets.
• The selected model, utilizing analysis of variance as the feature selector and linear discriminant analysis as the classifier, achieved AUCs of 0.838 (training), 0.814 (validation), and 0.824 (test).

Conclusions:

• FeAture Explorer (FAE) provides a convenient and effective tool for researchers to build and evaluate radiomics models.
• The software streamlines the process of model selection, comparison, and validation using independent test data.
• FAE is expected to be a valuable asset for radiomics studies and other medical research involving supervised machine learning, enhancing reproducibility and efficiency.