Differentiation of Brain Abscess From Cystic Glioma Using Conventional MRI Based on Deep Transfer Learning Features and Hand-Crafted Radiomics Features

Objectives: To develop and validate the model of combining deep transfer learning (DTL) features and hand-crafted radiomics (HCR) features for the differentiation of brain abscess from cystic glioma in conventional T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI). Methods: This single-center retrospective analysis included 188 patients with pathologically proven brain abscess or cystic glioma. 1000 DTL and 105 HCR features were extracted from the T1WI and T2WI of each patient. Cross-combination of three feature selection methods and four classifiers, such as k-nearest neighbors (KNN), random forest classifier (RFC), logistic regression (LR), and support vector machine (SVM), were compared to differentiate brain abscess from cystic glioma. The best method of combination of feature selection and classifier was chosen. Results: In most cases, deep learning-based radiomics (DLR) features, i.e., DTL features combined with HCR features, demonstrated higher diagnostic accuracy than HCR and DTL features alone for differentiating brain abscesses from cystic gliomas. The AUC of models based on DLR features in T2WI were 0.86 (95% CI: 0.81, 0.91) in the training cohort and 0.85 (95% CI: 0.75, 0.95) in the test cohort, respectively. Conclusion: Differentiation of brain abscess from cystic glioma can be predicted efficiently by the DLR model, providing a useful, inexpensive, convenient, and noninvasive strategy for differential diagnosis. More importantly, this is the first time that conventional MR imaging radiomics has been used to identify these diseases. Also, we have combined HCR and DTL features to get more impressive performance.