Due to the narrow time window for the treatment of acute ischemic stroke, the stroke lesion area in the patient must be identified as quickly and accurately as possible to evaluate the risks and get the most timely and effective treatment. Therefore, there is great clinical significance in the study of automatic identification and segmentation methods for stroke lesions. In this paper, we propose a partitioningstacking prediction fusion (PSPF) method based on an improved attention U-net to solve the problems of 3D-CNN-based networks, including their high computational cost and insufficient training data, and to achieve accurate segmentation of 3D stroke lesions. Our proposed PSPF method includes three steps in the first part. In Step 1, partitioning, we partition the slices obtained in a certain plane direction by slicing a Magnetic Resonance Imaging (MRI) into subsets according to the 2D graph similarity, then use each partitioned subset to perform training and prediction separately. In Step 2, stacking, we stack the 2D slice results of all subsets according to the position order in MRI before slicing and partitioning to form a 3D lesion result. In Step 3, fusion, we use soft voting to fuse the three orthogonal planes' 3D results that were obtained voxel by voxel in Steps 1 and 2. In the second part, we propose an improved attention U-net, which uses the features from three different scales to generate the attention gating coefficients that further improve training efficiency and segmentation accuracy. We implement a 6-fold cross-validation on the Anatomical Tracings of Lesions After Stroke (ATLAS) dataset to validate our method and model using metrics such as Dice Coefficient (DC), F2 score, and Area under Precision-Recall curve (APR). The results show that compared to the existing methods, our proposed method can not only improve the segmentation precision on unbalanced data but also improve the detailed performance of lesion segmentation. Our proposed method and model are generalized and accurate, demonstrating the good potential for clinical routines. The source codes and models in our method have been made publicly available at [available.upon.acceptance].

No abstract

For the segmentation task of stroke lesions, using the attention U-Net model based on the self-attention mechanism can suppress irrelevant regions in an input image while highlighting salient features useful for specific tasks. However, when the lesion is small and the lesion contour is blurred, attention U-Net may generate wrong attention coefficient maps, leading to incorrect segmentation results. To cope with this issue, we propose a dual-path attention compensation U-Net (DPAC-UNet) network, which consists of a primary network and auxiliary path network. Both networks are attention U-Net models and identical in structure. The primary path network is the core network that performs accurate lesion segmentation and outputting of the final segmentation result. The auxiliary path network generates auxiliary attention compensation coefficients and sends them to the primary path network to compensate for and correct possible attention coefficient errors. To realize the compensation mechanism of DPAC-UNet, we propose a weighted binary cross-entropy Tversky (WBCE-Tversky) loss to train the primary path network to achieve accurate segmentation and propose another compound loss function called tolerance loss to train the auxiliary path network to generate auxiliary compensation attention coefficient maps with expanded coverage area to perform compensate operations. We conducted segmentation experiments using the 239 MRI scans of the anatomical tracings of lesions after stroke (ATLAS) dataset to evaluate the performance and effectiveness of our method. The experimental results show that the DSC score of the proposed DPAC-UNet network is 6% higher than the single-path attention U-Net. It is also higher than the existing segmentation methods of the related literature. Therefore, our method demonstrates powerful abilities in the application of stroke lesion segmentation.

Deep learning-based techniques can obtain high precision for multimodal stroke segmentation tasks. However, the performance often requires a large number of training examples. Additionally, existing data extension approaches for the segmentation are less efficient in creating much more realistic images. To overcome these limitations, an unsupervised adversarial data augmentation mechanism (UTC-GAN) is developed to synthesize multimodal computed tomography (CT) brain scans. In our approach, the CT samples generation and cross-modality translation differentiation are accomplished simultaneously by integrating a Siamesed auto-encoder architecture into the generative adversarial network. In addition, a Gaussian mixture translation module is further proposed, which incorporates a translation loss to learn an intrinsic mapping between the latent space and the multimodal translation function. Finally, qualitative and quantitative experiments show that UTC-GAN significantly improves the generation ability. The stroke dataset enriched by the proposed model also provides a superior improvement in segmentation accuracy, compared with the performance of current competing unsupervised models.

Carotid artery stenosis is a serious medical condition that can lead to stroke. Using machine learning method to construct classifier model, carotid artery stenosis can be diagnosed with transcranial doppler data. We propose an improved fuzzy support vector machine model to predict carotid artery stenosis, with the maximum geometric mean as the optimization target. The fuzzy membership function is obtained by combining information entropy with the normalized class‐center distance. Experimental results showed that the proposed model was superior to the benchmark models in sensitivity and geometric mean criteria.