Detecting B-cell lymphoma-6 overexpression status in primary central nervous system lymphoma using multiparametric...

Area of Science:

• Neuro-oncology
• Radiology
• Machine Learning

Background:

• Primary central nervous system lymphoma (PCNSL) is a rare brain tumor.
• B-cell lymphoma-6 (BCL-6) is an unfavorable prognostic biomarker in PCNSL.
• Non-invasive detection of BCL-6 overexpression is crucial for patient management.

Purpose of the Study:

• To develop and validate a non-invasive method for detecting BCL-6 overexpression in PCNSL.
• Utilize multiparametric MRI and machine learning techniques for BCL-6 detection.
• Assess the diagnostic performance of radiomics-based models.

Main Methods:

• Retrospective analysis of 101 PCNSL lesions from 65 patients.
• Acquisition of multiparametric MRI data including DWI, T2WI, and T2FLAIR.
• Extraction of 2234 radiomics features, followed by LASSO selection.
• Development of machine learning models (LR, NB, SVM, KNN, MLP) for BCL-6 detection.
• Evaluation using sensitivity, specificity, accuracy, F1-score, and AUC.

Main Results:

• 30 radiomics-based models demonstrated varying degrees of BCL-6 status identification.
• Combining different MRI sequences and classifiers improved model performance.
• The Support Vector Machine (SVM) model, using three sequences, achieved the highest AUC of 0.95 in the training set and 0.87 in the validation set.

Conclusions:

• Multiparametric MRI combined with machine learning shows significant potential for non-invasively detecting BCL-6 overexpression in PCNSL.
• This approach could aid in prognostication and treatment planning for PCNSL patients.
• Further validation in larger cohorts is warranted.