TY - JOUR T1 - <strong>Prediction of IDH status and WHO grade in gliomas from 11C-Acetate PET/CT images using a convolutional neural network</strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3102 LP - 3102 VL - 63 IS - supplement 2 AU - Dongwoo Kim AU - Kyeong Taek Oh AU - Jong Hee Chang AU - Sun K Yoo AU - Mijin Yun Y1 - 2022/06/01 UR - http://jnm.snmjournals.org/content/63/supplement_2/3102.abstract N2 - 3102 Introduction: Glioma prognosis depends on WHO grade and isocitrate dehydrogenase (IDH) mutation status. We aimed to predict the WHO grade and IDH status of gliomas from preoperative 11C-Acetate PET images using convolutional neural networks (CNNs).Methods: 374 patient data were used to train the deep learning model. The two-dimentional axial slices were extracted from the patient volume data. From the extracted axial slices, 70% of the data were used as the training set, and the 30% of the data were used as the validation set. The axial slices were resized to 224X224 to match the input size of the deep learning model. For image augmentation, combination of image rotation, translation, scaling, and x-axis reflection was used. The architecture of resnet-50 was used to classify the type of brain tumor based solely on acetate-PET data. And 15 patients with gliomas were enrolled to test the CNN-based classifier for WHO grade and IDH status in prospective study. Results: Calculated performance for prediction of the IDH status was: 1 Slice (Accuracy 81.3%), 3 Slices (95.3%), 5 Slices (99.5%), 7 Slices (98.9%), 9 Slices (99.5%), and All Slices (99.3%) in validation set. And calculated performance for prediction of the WHO grade was: 1 Slice (Accuracy 76.8%), 3 Slices (97.9%), 5 Slices (98.0%), 7 Slices (98.6%), 9 Slices (99.3%), and All Slices (99.7%) in validation set. In prospectively enrolled 15 patients with gliomas, the accuracy of prediction of the IDH status was 93.3% (14/15), and the accuracy of prediction of the WHO grade was 96.7% (13/15).Conclusions: Our model demonstrated the high accuracy for the prediction of the IDH status, and WHO grade of gliomas in retrospective and also prospective patients. With deep learning network model, 11C-ACE PET/CT could play an important role as reading aid in the noninvasive prediction of the IDH status and WHO grade of gliomas. ER -