As an emerging biomedical image processing technology, medical image segmentation has made great contributions to sustainable medical care. Now it has become an important research direction in the field of computer vision. With the rapid development of deep learning, medical image processing based on deep convolutional neural networks has become a research hotspot. This paper focuses on the research of medical image segmentation based on deep learning. First, the basic ideas and characteristics of medical image segmentation based on deep learning are introduced. By explaining its research status and summarizing the three main methods of medical image segmentation and their own limitations, the future development direction is expanded. Based on the discussion of different pathological tissues and organs, the specificity between them and their classic segmentation algorithms are summarized. Despite the great achievements of medical image segmentation in recent years, medical image segmentation based on deep learning has still encountered difficulties in research. For example, the segmentation accuracy is not high, the number of medical images in the data set is small and the resolution is low. The inaccurate segmentation results are unable to meet the actual clinical requirements. Aiming at the above problems, a comprehensive review of current medical image segmentation methods based on deep learning is provided to help researchers solve existing problems.

The global outbreak of COVID-19 has become an important research topic in healthcare since 2019. RT-PCR is the main method for detecting COVID-19, but the long detection time is a problem. Therefore, the pathological study of COVID-19 with CT image is an important supplement to RT-RCT. The current TVLoss-based segmentation promotes the connectivity of diseased areas. However, normal pixels between some adjacent diseased areas are wrongly identified as diseased pixels. In addition, the proportion of diseased pixels in CT images is small, and the traditional BCE-based U-shaped network only focuses on the whole CT without diseased pixels, which leads to blurry border and low contrast in the predicted result. In this way, this paper proposes a SCTV-UNet to solve these problems. By combining spatial and channel attentions on the encoder, more visual layer information are obtained to recognize the normal pixels between adjacent diseased areas. By using the composite function DTVLoss that focuses on the pixels in the diseased area, the problem of blurry boundary and low contrast caused by the use of BCE in traditional U-shaped networks is solved. The experiment shows that the segmentation effect of the proposed SCTV-UNet has significantly improved by comparing with the SOTA COVID-19 segmentation networks, and can play an important role in the detection and research of clinical COVID-19.

Handwritten mathematical expression recognition (HMER) is crucial for intelligent human–machine interaction systems but remains challenging due to large variations in handwriting styles and the ambiguity of similar symbols. To address these challenges, we introduce WHATSNet, a framework for offline HMER. This framework combines wavelet transform techniques with advanced feature selection strategies. Unlike traditional spatial-domain methods, WHATSNet utilizes multiscale wavelet decomposition to extract style-invariant structural patterns. The model mitigates handwriting artifacts, such as noise and inconsistent symbol shapes, using adaptive frequency filtering. A hybrid attention mechanism is introduced to capture both fine-grained details and the global structural context. This mechanism is enhanced by a dual-query coverage attention mechanism, which enhances sequential prediction efficiency by incorporating historical context. Extensive evaluations on the CROHME benchmark datasets (2014, 2016, and 2019) show that WHATSNet outperforms state-of-the-art methods. The model achieves recognition accuracy improvements of 5.02%, 4.58%, and 5.07%, respectively. These results confirm the effectiveness of combining wavelet-based feature extraction with hybrid attention mechanisms for improving HMER performance. The code will be released at https://github.com/runminwang/WHATSNet.

Temporal action localization (TAL) is a research hotspot in video understanding, which aims to locate and classify actions in videos. However, existing methods have difficulties in capturing long-term actions due to focusing on local temporal information, which leads to poor performance in localizing long-term temporal sequences. In addition, most methods ignore the boundary importance for action instances, resulting in inaccurate localized boundaries. To address these issues, this paper proposes a state space model for temporal action localization, called Separated Bidirectional Mamba (SBM), which innovatively understands frame changes from the perspective of state transformation. It adapts to different sequence lengths and incorporates state information from the forward and backward for each frame through forward Mamba and backward Mamba to obtain more comprehensive action representations, enhancing modeling capabilities for long-term temporal sequences. Moreover, this paper designs a Boundary Correction Strategy (BCS). It calculates the contribution of each frame to action instances based on the pre-localized results, then adjusts weights of frames in boundary regression to ensure the boundaries are shifted towards the frames with higher contributions, leading to more accurate boundaries. To demonstrate the effectiveness of the proposed method, this paper reports mean Average Precision (mAP) under temporal Intersection over Union (tIoU) thresholds on four challenging benchmarks: THUMOS13, ActivityNet-1.3, HACS, and FineAction, where the proposed method achieves mAPs of 73.7%, 42.0%, 45.2%, and 29.1%, respectively, surpassing the state-of-the-art approaches.