Application of Deep Learning on MRI for Discriminating Glioma Recurrence from Radiation Necrosis: Algorithm Development and Validation

Purpose: Accurate differentiation between glioma recurrence and radiation necrosis is crucial for the management of patients suspected of glioma recurrence after radiation therapy. This study aims to develop a deep learning methodology for automated discrimination of glioma recurrence and radiation necrosis using routine magnetic resonance imaging (MRI) scans. Method: We investigated 234 patients who underwent radiotherapy following glioma resection and presented with suspected recurrent lesions during follow-up MRI examinations retrospectively. Routine 3D-MRI scans, including T1, T2, and T1ce sequences, were obtained for each patient. Out of the analyzed cases, 192 (82.1%) were pathologically confirmed as glioma recurrence, while 46 (16.1%) were diagnosed as radiation necrosis. Different Convolutional Neural Network (CNN) models were utilized to learn radiological features indicative of gliomas and necrosis from the MRI scans. Performance evaluation metrics including sensitivity, specificity, accuracy, and area under the curve (AUC), were employed to assess the models’ performance. Result: Among the evaluated CNN models, ResNet10 exhibited the highest sensitivity of 0.778, specificity of 0.939, accuracy of 0.914, and an AUC of 0.828. Additionally, the MresNet model achieved the highest specificity of 0.980 but had a lower sensitivity of 0.556. Another evaluated CNN model, Vgg16, displayed a sensitivity of 0.556, specificity of 0.939, accuracy of 0.879, and an AUC of 0.702. Conclusion: The proposed ResNet10 CNN model shows promising performance on routine MRI scans, making it highly applicable in clinical settings. These findings contribute to improving the diagnostic accuracy in distinguishing between glioma recurrence and radiation necrosis based on routine MRI.