Colorectal Cancer Detection Using Deep and Transfer Learning

ABSTRACT: Early correct diagnosis of Colorectal cancer stands as the fundamental factor to boost patient survival because this cancer ranks as a global leader both in incidence and mortality rates. The standard colonoscopy with biopsy tests work well in diagnosis yet they severely impact patient comfort through their extensive nature and variable observational results. The recent developments in artificial intelligence particularly deep learning enabled precise automated interpretation of medical images through advances made in recent years. A study researches the identification of colorectal cancer in histopathological images while using Xception and MaxVit networks with transfer learning. The architectures were chosen considering their established strengths the depthwise separable convolutions of Xception that facilitate efficiency and the hybrid vision transformer architecture of MaxViT that captures local and global image dependencies. 1,51,118 histopathological images from a dataset were classified into Microsatellite Stable (MSS) and Microsatellite Instability-Mutated (MSIMUT) classes for training, validation, and testing. The system demonstrated strong performance in model classification after pre-training its versions while reducing expenses and accelerating training duration. The proposed approach strengthens pathologic assessments through reliable outcomes and stable performance while introducing scalability to help clinicians with diagnosis processes for CRC in an expedient manner. IndexTerms -Colorectal Cancer (CRC), Deep Learning, Transfer Learning, Histopathology /Histopathological Images, Microsatellite Stable (MSS), Microsatellite Instability-Mutated (MSIMUT), Xception Model.