Adaptive Attention Convolutional Neural Network for Liver Tumor Segmentation

Abstract

PurposeAccurate segmentation of liver and liver tumors is critical for radiotherapy. Liver tumor segmentation, however, remains a difficult and relevant problem in the field of medical image processing because of the various factors like complex and variable location, size, and shape of liver tumors, low contrast between tumors and normal tissues, and blurred or difficult-to-define lesion boundaries. In this paper, we proposed a neural network (S-Net) that can incorporate attention mechanisms to end-to-end segmentation of liver tumors from CT images.MethodsFirst, this study adopted a classical coding-decoding structure to realize end-to-end segmentation. Next, we introduced an attention mechanism between the contraction path and the expansion path so that the network could encode a longer range of semantic information in the local features and find the corresponding relationship between different channels. Then, we introduced long-hop connections between the layers of the contraction path and the expansion path, so that the semantic information extracted in both paths could be fused. Finally, the application of closed operation was used to dissipate the narrow interruptions and long, thin divide. This eliminated small cavities and produced a noise reduction effect.ResultsIn this paper, we used the MICCAI 2017 liver tumor segmentation (LiTS) challenge dataset, 3DIRCADb dataset and doctors’ manual contours of Hubei Cancer Hospital dataset to test the network architecture. We calculated the Dice Global (DG) score, Dice per Case (DC) score, volumetric overlap error (VOE), average symmetric surface distance (ASSD), and root mean square error (RMSE) to evaluate the accuracy of the architecture for liver tumor segmentation. The segmentation DG for tumor was found to be 0.7555, DC was 0.613, VOE was 0.413, ASSD was 1.186 and RMSE was 1.804. For a small tumor, DG was 0.3246 and DC was 0.3082. For a large tumor, DG was 0.7819 and DC was 0.7632.ConclusionS-Net obtained more semantic information with the introduction of an attention mechanism and long jump connection. Experimental results showed that this method effectively improved the effect of tumor recognition in CT images and could be applied to assist doctors in clinical treatment.

Accurate and robust auto‐segmentation of head and neck organ‐at‐risks based on a novel CNN fine‐tuning workflow

“…For more details on HN-Net and the attention model we adopted, please refer to our previous work. 34,35 In the fine-tuning process, we selected the top 30% of the uncertainty images to retrain the model. When we selected the top 60% or 90% of uncertainty images, it is found that the DSC values showed a minor performance improvement, while the training time escalated significantly.…”

“…CBAM is able to infer from two different dimensions (spatial and channel dimensions) for the same input, leading to improved feature aggregation and semantic reasoning. For more details on HN‐Net and the attention model we adopted, please refer to our previous work 34,35 …”

“…fine‐tuned segment‐anything‐model (SAM) for medical image segmentation through freezing the image encoder and prompt encoder and only fine‐tune the mask decoder 38 . In recent publications, we and co‐authors used contrastive learning strategy to pre‐train encoder weights on public datasets and transfer them to local tasks for fine‐tuning 35 . We lack specific network architectures and training information for these commercial AI software, and we also don't know whether they initially employed fine‐tuning for transfer learning.…”

A multi-modal vision-language pipeline strategy for contour quality assurance and adaptive optimization

“…In recent years, deep learning has provided promising solutions for many data-driven clinical challenges (Protonotarios et al 2022, Wang et al 2022a, Luan et al 2021. As a result, an increasing number of deep learning-based commercial artificial intelligence software (DL-CAIS) are being introduced into clinical practice.…”

Deep learning for head and neck semi-supervised semantic segmentation

“…Traditionally, RT planning involves manual delineation of target and OARs on CT images, which necessitates the expertise of physicians, a process that is both labor-intensive and time-consuming (Xu et al 2022). Nowadays, computer-assisted medical diagnosis has evolved into a prevalent tool, enhancing the capabilities of lesion detection, diagnosis, and segmentation (Luan et al 2021, Zhang et al 2021a, Luan et al 2023a. Despite the comprehensive research conducted on automatic H&N segmentation, the absence of precise delineation for H&N OARs and the limitation of using single dataset in model training have contributed to deficiencies in the robustness of automatic segmentation algorithms (Gao et al 2019, Liang et al 2020.…”