Hippocampal And Subfields Segmentation Algorithm Based On Deep Learning Model

Hippocampus is a structure in the brain in memory consolidation and can be divided into nine subfields.Hippocampus atrophy has been mostly studied in various neurological diseases,such as Alzheimer’s disease and mild cognitive impairment.Accurate hippocampus subfields segmentation in magnetic resonance(MR)images plays a crucial role in the diagnosis,prevention,and treatment of neurological diseases.However,the segmentation is a challenging task due to small size,relatively low contrast,complex shape,and indistinct boundaries of hippocampus subfields.manual segmentation is still considered the gold standard.However,manual segmentation is time-consuming and labor-intensive and thus often not feasible for large data sets.Therefore,there is a strong need to have efficient computer based automatic segmentation method.Multi-atlas based methods can obtain accurate segmentation results by fusing propagated labels of multiple atlases in a target image space.However,the performance of multi-atlas significantly relies on the effectiveness of the label fusion method,and the hand-crafted feature used by traditional method is not sufficiently for the segmentation task.Inspired by the recent success of deep learning models in solving various vision task,such as classification and segmentation.this thesis proposes a deep learning models based hippocampus and subfields segmentation method.The following three aspects are studied in detail:(1)Deep learning model based image classification and segmentation task is studied.First,we discuss the different segmentation method using the convolutional neural network and stack convolutional autoencoder,by which two type of segmentation method can be obtained: pixel-classification based method and semantic based method.Then,we describe the advantage and disadvantage between this two different networks and segmentation methods.Finally,we introduce this two deep learning models into the field of hippocampus segmentation.(2)Based on semantic segmentation method,a U-Res-Net which Combined the residual block with the stack convolutional autoencoder is designed to segment the whole hippocampus.Also,in order to improve the performance of the U-Res-Net,the adversarial training algorithm is introduced for the training of the U-Res-Net,the experiment show that U-Res-Net with adversarial training outperform the traditional method.(3)In the process of segmenting the subfields of hippocampus,compared with the hippocampus segmentation,we first discuss the difficulties of the subfields segmentation,then,we analyze the reasons why the subfields can not be effectively segmented using the semantic segmentation.In order to overcome these limitations,we propose a combined multi-scale patch and cascaded CNN-based method for hippocampus subfields segmentation.Two different CNN,Big Sub Net and Small Sub Net,is designed for big and small subfields.The significant improvement compared with traditional method shows that the proposed algorithm can label with high accuracy and is highly appropriate for hippocampus subfield segmentation.