Automated classification of the phasecase scale in mri using deep learning methods

The aim of this study was to classify the Fazekas scale, which indicates the burden of periventricular white matter hyperintensities (WMH), especially those caused by chronic small vessel disease, using deep learning techniques on MRI. Material and Methods: A total of 1266 patients (590 males and 676 females, aged 18-96 years) who underwent brain MRI at Van Yüzüncü Yıl University Dursun Odabaşı Medical Center Radiology Clinic between December 2023 and April 2024 and whose images were normal and BCH was detected in axial fat-suppressed FLAIR WI sequences were retrospectively reviewed. Images were classified into 4 classes according to Fazekas scale. For classification, we performed feature extraction on EfficientNet, one of the deep learning models, and then classified with SVM, KNN and XGBoostClassifier classifiers with CV and LOOCV crossover techniques and compared their performances. Results: Feature extraction was performed on EfficientNet. Among the extracted features, 100 features were selected with PCA and the results were obtained by running 3 classification algorithms initially. The best result was obtained from the XGBoostClassifier classifier with the features extracted from the EfficientNet B1 model with a high rate of approximately 95%. In addition, performance tests were also performed with 8 classification algorithms using CV and LOOCV crossover techniques on the extracted features. In these performance tests, an accuracy rate of 95% was achieved on the features obtained from the Efficient B1 model. Conclusion: Evaluating deep learning architectures for faster and easier detection of WMHs that can cause serious health problems can significantly reduce the workload of radiology physicians by providing ease of diagnosis.