Multi-plane attention guided 3D nnU-Net for MRI-based cervical cancer segmentation.

Objective: Over 300,000 people die from cervical cancer, which is the fourth most frequent malignancy in women worldwide. Early identification of cervical cancer has been related to much greater survival rates, and the illness is usually preventable. Accurate cervical segmentation using magnetic resonance imaging (MRI) is critical for diagnosis, treatment planning and response assessment especially in image-guided brachytherapy. Methods: In terms of 3D MRI cervical tumor image segmentation task, several cutting-edge state-of-the-art architectures are explored. We propose a novel model having MultiPlane Attention Guided Enhanced nn-UNet3D model that uses volumetric contextual information to learn spatial relationships across slices simultaneously improving border localization and noise robustness. Results: It is applied on publicly available TCGA-CESC dataset which is part of the Cancer Genome Atlas program (TCGA) and it focuses on cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC). Preliminary results show the accuracy, IoU and Dice score values of 95.04%, 87.87% and 93.52% respectively. Conclusion: This study discusses the significance of MRI scans, which provide high-resolution anatomical information, improving the accuracy of segmentation algorithms in identifying and characterizing aberrant cervical tissues. Future research will concentrate on domain generality across scanners and institutions, multimodal MRI integration and clinical validation in prospective therapy scenarios.