Differentiation of Enhancing Glioma and Primary Central Nervous System Lymphoma by Texture-Based Machine Learning

Area of Science:

• Neuroimaging
• Machine Learning
• Oncology

Background:

• Differentiating primary central nervous system lymphoma (PCNSL) from enhancing gliomas is crucial for treatment and avoiding unnecessary surgery.
• Accurate preoperative diagnosis is essential for optimal patient management.

Purpose of the Study:

• To evaluate the diagnostic performance of a machine-learning algorithm using texture analysis of contrast-enhanced T1-weighted images (CE-T1WI).
• To differentiate between PCNSL and enhancing gliomas.

Main Methods:

• Texture analysis of CE-T1WI was performed on 71 enhancing gliomas and 35 PCNSLs.
• A support vector machine (SVM) classifier was trained using textural features.
• Diagnostic accuracy was compared to three expert readers using areas under the receiver operating characteristic curves (AUC).

Main Results:

• The SVM classifier achieved a mean AUC of 0.877.
• The SVM classifier's performance was noninferior to the mean AUCs of the three readers (0.878, 0.899, 0.845).
• Statistical analysis confirmed the noninferiority of the SVM classifier (P = .021, P = .035, P = .007).

Conclusions:

• SVM classification based on CE-T1WI textural features is a reliable tool for differentiating PCNSL from enhancing glioma.
• This AI approach demonstrates noninferior diagnostic accuracy compared to expert neuroradiologists.
• Machine learning offers a promising non-invasive method for improved preoperative diagnosis in neuro-oncology.