Evaluation of brain parenchyma magnetic resonance imaging findings in gaucher disease and comparison of differences between subtypes of the disease with radiomics analysis

Objective: Gaucher Disease is the most common lysosomal storage disease and is divided into 3 clinical subtypes. In this thesis, it was aimed to investigate the detectability of differences between Gaucher Disease subtypes and cerebral involvement by radiomics analyzes performed on cerebral magnetic resonance images. Materials and Methods: Cerebral magnetic resonance imaging of 25 type 1 and 26 type 3 Gaucher patients, whose follow-up and treatment was carried out by Çukurova University Faculty of Medicine, Department of Child Nutrition and Metabolism, obtained between March 2016 and January 2023 in Cukurova University Faculty of Medicine Department of Radiology and in different centers in the surrounding provinces images were included in the study retrospectively. After segmentation, radiomics data was obtained by feature extraction after segmentation from the left putamen on T2-weighted axial images, left hippocampus and left posterior peritrigonal white matter on FLAIR images, and a total of 333 quantitative parameters were obtained for each patient. After feature reduction with Gini index, the 5 most significant parameters were determined. Analyzes made with 5 different machine learning methods were compared with parameters such as area under the curve, accuracy, sensitivity, specificity, F1 score, precision, operating characteristic curve and error/confusion matrix. Results: A total of 51 patients aged between 3-62 years were included in the study. Twenty-four (47.1%) cases were female and 27 (52.9%) were male. The mean age of the cases was 21.2±13.7 years. Twenty-five (49%) of the cases were type 1 and 26 (51%) were type 3 Gaucher patients. There was no significant difference between type 1 and type 3 Gaucher patient groups according to age and gender. (p>0.05) Three of the 5 parameters obtained after feature reduction belonged to hippocampal radiomics analyses. The highest success in modeling with machine learning methods was obtained in the neural network model with 0,808 AUC, 0,725 accuracy rate, 0,727 F1 score, 0,727 positive prediction, 0,725 sensitivity, and 0,750 specificity values. The neural network algorithm was able to correctly classify 76% (n=19) of 25 type 1 Gaucher patients and 69% (n=18) of 26 type 3 Gaucher patients. In addition, as a result of analysis of variance, hippocampal involvement was noted in type 1 Gaucher patients. Conclusion: According to the results of our study, radiomics analyzes performed on cerebral magnetic resonance imaging can differentiate between Gaucher disease subtypes with high success. According to the comparisons made between the patient groups, the most significant features were obtained from the hippocampus level, and it is thought that the hippocampus is the localization of selective involvement for the disease.