CT Image Segmentation Of Intracranial Hemorrhage Based On Deep Convolutional Networks

Intracerebral hemorrhage(ICH)is a serious category of head injury,which means bleeding occurs inside the skull,and in many cases leads to a high disability or mortality and threatens the lives of patients.Therefore,accurate and rapid hematoma region segmentation is of great significance for ICH diagnosis and treatment.In view of the fact that traditional methods cannot fully automate the segmentation of lesions and that existing methods still have limitations for the accurate segmentation of small targets or irregularly shaped hemorrhage regions,this paper provides an in-depth study of the segmentation method of lesion regions based on deep learning according to the characteristics of CT images of intracranial haemorrhage.Aiming at the problem that the bleeding area is small and difficult to segment in early intracranial hemorrhage CT images,an end-to-end network model named IHA-Net is proposed.Firstly,in order to prevent the degradation of tiny lesion regions due to repeated down-sampling,we proposed a Residual Hybrid Atrous Convolution strategy.Furthermore,a multi-object function for joint optimization is introduced to ease the severe pixel imbalanced problem.In addition,to avoid gradient disappearance and slow convergence,intermediate supervision is utilized during the training and an attention mechanism is added to the decoder stage.Through quantitative experimental comparisons based on Dice,Jaccard,Sensitivity and other assessment metrics,qualitative visualization results and clinical trial results,it is verified that the IHA-Net model proposed in this paper has high feasibility and clinical application for early intracranial hemorrhage CT image segmentation.Aiming at the problems of different shapes and unsmooth boundaries of bleeding areas in CT images of intracranial hemorrhage,a novel network framework named TACL-Net is proposed.Firstly,a new loss function is proposed,which is different from the traditional loss function,combining the active contour theory to give more attention to the geometric contour information of the lesion area.Secondly,a segmentation network model incorporating an improved attention mechanism is proposed.Experiments are conducted on a real clinical intracranial hemorrhage CT image dataset,and verifies that the TACL-Net model proposed in this paper shows greater advantages in the intracranial haemorrhage CT image segmentation task through experimental comparison of quantitative metrics and qualitative visualisation with other classical models.In addition,based on the segmentation of intracranial haemorrhage lesions,we also focused on the performance of estimating haematoma volume in intracranial haemorrhage for the test case and used the Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)as evaluation metrics to compare with other models in terms of volume estimation,achieving better results.