Metabolic index of the best preserved hemisphere of 18F-FDG PET imaging combination behavioral CRS-R scores predict recovery from disorders of consciousness

Abstract Background The objective of this study was to develop a prognostic model for predicting one-year post-injury outcomes in chronic disorders of consciousness (DoC) by detecting relatively preserved brain metabolism through ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG PET). This approach allows the assessment of the level of consciousness and the prediction of the likelihood of wakefulness. Methods Eighty-seven DoC patients newly diagnosed with behavioral Revised Coma Recovery Scale (CRS-R) and ¹⁸F-FDG PET/CT studies were included. PET images were standardized by the metabolic index of the best-preserved hemisphere (MIBH) and the ratio SUV (SUVR), respectively. The training of image-based classification was conducted using the DenseNet121 network, while tabular-based deep learning was employed for training depth features extracted from imaging models and behavioral CRS-R scores. The performance of the models was assessed using the area under the curve (AUC). Results Of the 87 DoC patients who received routine treatments, consciousness recovery was observed in 52 patients, while consciousness non-recovery was observed in 35 patients. The classification performance of the MIBH model was found to be superior to that of the SUVR model, with AUC values of 0.751 ± 0.093 and 0.412 ± 0.104 on the test sets, respectively. The MIBH + CT multimodal model was determined to perform better than the MIBH-only model, achieving an AUC of 0.784 ± 0.073 on the test sets. The combination of MIBH + CT depth features with behavioral CRS-R scores resulted in the best classification accuracy, with AUC values of 0.950 ± 0.027 and 0.933 ± 0.015 on the training and test sets, respectively. Conclusions The prediction of recovery in DoCs was facilitated by a model based on a combination of multimodal imaging features and behavioral CRS-R scores.