Опубликована: Авг. 7, 2024
Язык: Английский
Medical Physics, Год журнала: 2025, Номер 52(5), С. 3135 - 3150
Опубликована: Фев. 3, 2025
Abstract Background Most attention‐based networks fall short in effectively integrating spatial and channel‐wise information across different scales, which results suboptimal performance for segmenting coronary vessels x‐ray digital subtraction angiography (DSA) images. This limitation becomes particularly evident when attempting to identify tiny sub‐branches. Purpose To address this limitation, a multi‐scale dual attention embedded network (named MDA‐Net) is proposed consolidate contextual channel contiguous levels scales. Methods MDA‐Net employs five cascaded double‐convolution blocks within its encoder adeptly extract features. It incorporates skip connections that facilitate the retention of low‐level feature details throughout decoding phase, thereby enhancing reconstruction detailed image information. Furthermore, MDA modules, take features from neighboring scales hierarchical levels, are tasked with discerning subtle distinctions between foreground elements, such as diverse morphologies dimensions, complex background, includes structures like catheters or other tissues analogous intensities. sharpen segmentation accuracy, utilizes composite loss function integrates intersection over union (IoU) binary cross‐entropy loss, ensuring precision outcomes maintaining an equilibrium positive negative classifications. Results Experimental demonstrate not only performs more robustly on DSA images under various conditions, but also achieves significant advantages state‐of‐the‐art methods, achieving optimal scores terms IoU, Dice, Hausdorff distance 95%. Conclusions has high robustness segmentation, providing active strategy early diagnosis cardiovascular diseases. The code publicly available at https://github.com/30410B/MDA‐Net.git .
Язык: Английский
Процитировано
0
Computerized Medical Imaging and Graphics, Год журнала: 2025, Номер unknown, С. 102540 - 102540
Опубликована: Март 1, 2025
Язык: Английский
Процитировано
0
Journal of Radiation Research and Applied Sciences, Год журнала: 2023, Номер 16(4), С. 100680 - 100680
Опубликована: Окт. 3, 2023
Transformers perform well in natural language processing tasks and have made many breakthroughs computer vision. In medical image processing, transformers are successfully used segmentation, classification, reconstruction, diagnosis. this paper, we mainly expound on the transformer principle its application imaging. Specifically, first introduce basic principles model structure of transformers. Then, summarize improvement mechanism transformer's network including combining with Unet network, creating a lightweight variant strengthening cross-fast link mechanism, building large as skeleton. Second, extensive discussion is given to other applications. Finally, main challenges face field future development prospects. Furthermore, systematically latest research progress their which has significant reference value for field.
Язык: Английский
Процитировано
7
Computerized Medical Imaging and Graphics, Год журнала: 2024, Номер 117, С. 102426 - 102426
Опубликована: Авг. 31, 2024
Язык: Английский
Процитировано
1
Computer Methods and Programs in Biomedicine, Год журнала: 2024, Номер 258, С. 108463 - 108463
Опубликована: Окт. 22, 2024
Язык: Английский
Процитировано
1
Heliyon, Год журнала: 2024, Номер 10(22), С. e38579 - e38579
Опубликована: Сен. 27, 2024
Язык: Английский
Процитировано
0
Опубликована: Авг. 7, 2024
Язык: Английский
Процитировано
0