Research On Brain Tumor Image Segmentation Algorithm Based On Fully Convolutional Neural Network

Brain tumor is a common neurological disease,which greatly endangers human health and even life-threatening.Accurately extracting brain tumor information based on Magnetic Resonance Imaging(MRI)and managing and summarizing the captured information to segment brain tumors are critical to assist in the diagnosis of brain tumors.However,manual delineation is not only time-consuming and laborious,but also error prone.The different phenotypes,sizes and locations of brain tumors make automatic segmentation a challenging task.In recent years,deep learning models based on artificial neural networks have shown advanced performance in the field of medical image processing.Convolutional neural networks have gradually attracted attention in MRI-based brain tumor segmentation tasks with their powerful feature extraction capabilities.In order to accurately and automatically segment brain tumors,this paper proposes two automatic segmentation algorithms based on fully convolutional neural networks,which are dedicated to the segmentation of multiple sub-regions connected to each other and the segmentation of multiple objects not connected to each other,respectively.The two algorithms are applied to glioma image segmentation and multiple brain metastasis image segmentation.The main works are as follows:(1)The different phenotypes,sizes and locations of gliomas in/between patients and the fuzzy boundaries of gliomas make automatic segmentation of all diverse gliomas,including the whole tumors(WTs),tumor cores(TCs)and enhancing tumors(ETs)of high-grade gliomas(HGG)and low-grade gliomas(LGG)a challenging task.To alleviate these challenges,in this paper,we propose a 3D hierarchical fully convolutional network(FCN)with a dual-attention(i.e.,global and local attention)mechanism to segment diverse gliomas simultaneously.The global attention mechanism(GAM)focuses on segmenting gliomas precisely by segment discrimination learning with a weight-allocated segmentation loss function to alleviate biased results obtained for tumors with large sizes and an adversarial loss function to refine the segmentations of areas with low contrast relative to their neighbors.The local atte ntion mechanism(LAM)constantly revises effective features with the guidance of a united loss function at different levels.Furthermore,we present a hierarchical feature module(HFM)with a weight-sharing block to obtain more information about the boundaries of different scales,aiming at enhancing the learning of ambiguous tumor outlines.According to experimental results,our network outperforms ten state-of-the-art methods.Ablation studies show that the proposed model components are effective for diverse glioma segmentation.(2)Brain metastasis refers to the metastasis of tumor cells from other parts of the body into the brain,which has a high fatality rate.The size,location and number of multiple brain metastases within and between patients are different,which makes it difficult to accurately and automatically segment multiple brain metastases.This paper proposes a single-pass fully convolutional network(SPnet)based on an encoding-decoding framework for end-to-end automatic segmentation of multiple brain metastases.SPnet is a single-stage deep learning model with a multilevel fusion attention mechanism and multiresidual connection modules.It can achieve pixel-level segmentation of multiple brain metastases based on MRI axial slices.The multilevel fusion attention mechanism reduces the interference from redundant information from similar tissues around the brain metastases and thereby reduces misidentification.The multiresidual connection modules can fuse features of different depths in the ne twork and improve feature utilization.Experimental results prove that the model proposed in this paper is superior to the existing methods,and can achieve satisfactory segmentation results in brain metastases of various sizes and patients with different numbers of brain metastases.This model shows good performance in particularly small brain metastases(<3 mm)and patients with a large number of tumors(>10).Furthermore,ablation studies verify the effectiveness of the proposed modules for the segmentation of multiple brain metastases.