Purpose: Knowing the course of Alzheimer's disease is very important to prevent the deterioration of the disease, and accurate segmentation of sensitive lesions can provide a visual basis for the diagnosis results. This study proposes an improved end-to-end dual-functional 3D convolutional neural network for segmenting bilateral hippocampi from 3D brain MRI scans and diagnosing AD progression states simultaneously.
Methods: The proposed neural network is based on the V-Net and adopts an end-to-end structure. In order to relieve the excessive amount of convolutional parameters at the bottom of the V-Net, we change them to bottleneck architecture. Based on the segmentation network, we establish a classification network for diagnosing pathological states of brain. In order to balance the two tasks of hippocampi segmentation and brain pathological states diagnosis, we designed a unique loss function. This study included 132 samples, of which 100 were selected as training, and the remaining 32 were used to test the performance of our model. During training, we adopted fivefold cross-validation method.
Results: We selected the intersection over union and dice coefficient to evaluate the hippocampus segmentation performance, while the brain pathological states diagnosis performance was evaluated by accuracy, specificity, sensitivity, precision and F1 score. By using the proposed neural network, the left hippocampi segmentation Iou and dice coefficient reach 0.8240 ± 0.020 and 0.9035 ± 0.020, respectively. The right hippocampi Iou and dice coefficient reach 0.8454 ± 0.023 and 0.9162 ± 0.023, respectively. The accuracy, specificity, sensitivity, precision and F1 score of three-category classification of brain pathology are 84%, 92%, 84%, 86% and 85%, respectively.
Conclusion: The proposed neural network has two functions of brain pathological states diagnosis and bilateral hippocampi segmentation with higher robustness and accuracy, respectively. The segmented bilateral hippocampi can be used as a reference for clinical decision making or intervention.
Keywords: Alzheimer’s disease; Classification; Computer-aided diagnosis; Convolutional neural network; Segmentation.