Diffusion-weighted imaging-based radiomics model using automatic machine learning to differentiate cerebral cystic...

Area of Science:

• Radiology
• Machine Learning
• Oncology

Background:

• Cerebral cystic metastases and brain abscesses present similar imaging characteristics.
• Accurate differentiation is crucial for appropriate patient management and treatment.

Purpose of the Study:

• To develop and validate a radiomics model using diffusion-weighted imaging (DWI) to distinguish between cerebral cystic metastases and brain abscesses.
• To leverage automated machine learning for optimizing the diagnostic model.

Main Methods:

• Retrospective analysis of 186 patients (98 with cystic metastases, 88 with abscesses) from two institutions.
• Radiomics features extracted from DWI images (cystic core and solid wall) after robust preprocessing.
• Tree-based Pipeline Optimization Tool (TPOT) used for automated machine learning model selection and optimization with cross-validation.

Main Results:

• An optimized TPOT model, utilizing radiomics signatures from both lesion subregions and wavelet transform, achieved high performance.
• Internal testing: AUC of 1.00, accuracy 0.97, sensitivity 1.00, specificity 0.93.
• External validation: AUC of 1.00, accuracy 0.96, sensitivity 1.00, specificity 0.93.

Conclusions:

• The developed DWI-based radiomics model demonstrates significant potential for accurately differentiating cerebral cystic metastases from brain abscesses.
• Automated machine learning (TPOT) effectively optimized the radiomics pipeline for robust diagnostic performance.