Can 3D T1 Post-Contrast MRI in A Radiomics-Machine Learning Model Distinguish Infective from Neoplastic...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate classification of ring-enhancing brain lesions (REBLs) into infection or neoplasm is crucial for timely clinical management.
• Radiomics analysis of high-resolution 3D T1 post-contrast MRI data offers potential for characterizing brain lesions.
• The utility of radiomics in differentiating central nervous system infections remains underexplored.

Purpose of the Study:

• To develop and evaluate a radiomics-machine learning model for distinguishing infective from neoplastic REBLs.
• To assess the model's performance using only 3D T1 post-contrast MRI data.
• To compare the radiomics-machine learning model with 2D CNN and hybrid approaches.

Main Methods:

• A dataset of 92 patients with 402 REBLs was used for training and validation.
• 1197 radiomics features were extracted, followed by feature selection and application of nine machine learning classifiers.
• Model performance was validated on an external holdout dataset of 57 patients with 454 REBLs.

Main Results:

• The Multi-layer Perceptron (MLP) model achieved a mean AUC of 0.80 in cross-validation and 0.84 on the external holdout dataset.
• The model demonstrated robust performance with sensitivity of 0.84 and specificity of 0.75 on external data.
• Comparable performance was observed between the radiomics-MLP model and CNN-based approaches.

Conclusions:

• A radiomics-machine learning model based solely on 3D T1 post-contrast MRI shows promise in differentiating infective from neoplastic REBLs.
• The model's stable performance on external data suggests its potential clinical utility.
• Further research incorporating multimodal MRI sequences and clinical data is recommended for enhanced diagnostic accuracy.