Enhancing Lymph Node Classification in Rectal Cancer MRI Images Through a Multi-Model Approach

Rectal cancer remains to be a major issue worldwide, necessitating accurate and precise methods for proper classification of lymph nodes in MRI scan images to guide treatment decisions. The use of deep learning based techniques can help with the task of classification. This research focuses on enhancing the process of lymph node classification in rectal cancer MRI images by using a multi-model approach. Firstly, we explore the use of diffusion models in data augmentation, to produce synthetic lymph node images to counter the issue of limited dataset size. Secondly, Real-ESRGAN and Swin2SR super-resolution techniques are examined for their ability to improve image quality and hence enhance the accuracy of lymph node classifications. Lastly, ConvNeXt is employed as an advanced technique used to improve overall performance at classifying lymph nodes for rectal cancer. Through a series of detailed experiments on the dataset consisting of T2-Weighted Axial MRI scans from 54 patients, we demonstrate the effectiveness of our approach in classification of lymph nodes. The metrics like accuracy and precision exhibit significant improvements, highlighting the superiority of our multi-model approach.