MBGCU: A precise segmentation network for the levator ani muscle and surrounding tissues

The edema of levator ani muscle and surrounding tissues is a common manifestation of muscle injury. Accurately identifying it is crucial for early prevention and clinical intervention of pelvic organ prolapse after childbirth. In MRI images, the boundaries of the puborectalis muscle region are often unclear, and its characteristics are similar to those of surrounding soft tissues, making it challenging to assess edema precisely. This study presents a deep learning model, MBGCU, to address this challenge for segmenting the edema of levator ani muscle and surrounding tissues based on the MAMBA framework. The model optimizes multi-scale feature extraction through the VSSM module, improving detection in regions with unclear boundaries. The innovative GCUP module dynamically adjusts the feature neighborhood relationships to capture local geometric information more accurately. The SU fusion upsampling module enhances feature expression precision, improving segmentation performance. Experimental results show that on the pelvic floor MRI dataset, the MBGCU model achieves a Dice coefficient of 0.8048 and mIoU of 0.8402, significantly outperforming existing state-of-the-art models. This model demonstrates excellent accuracy and stability in identifying the edema of levator ani muscle and surrounding tissues and provides strong data support and technical assurance for early prevention and clinical intervention of POP. • A novel MBGCU is developed for segmenting levator ani muscle and surrounding tissues. • The VSSM module captures multi-scale, blurred features for better lesion detection. • The GCUP and SU modules enhance features in low-contrast pelvic MRI regions. • MBGCU achieves Dice 0.8048 and mIoU 0.8402, outperforming others on benchmarks.