Enhanced Chromosome Defect Detection Using Hybrid Swin Transformer and Explainable AI

This research offers a new method for detecting chromosome defects using a Hybrid Swin Transformer integrated with Explainable AI (XAI), incorporating advanced algorithms for enhanced performance. The Hybrid Swin Transformer leverages hierarchical feature learning for robust feature extraction, while XAI provides interpretability and transparency in defect classification, achieving an impressive accuracy of 97.8%, surpassing existing solutions (85%-96.5%). The architecture incorporates Neural Image Segmentation (UNet-based) for segmenting chromosomes from karyotype images, achieving a segmentation accuracy of 95.1%, significantly enhancing defect region identification. Data augmentation and preprocessing techniques, such as normalization and G-banding enhancement, ensure the model effectively generalizes across diverse datasets. The workflow is further optimized with XAI for adaptability to new classes of abnormalities and decision transparency, reducing the 2.5% for false positives and 3.4% for false negatives. rateof false negatives (3.4%) and false positives (2.5%). The architecture processes images in an average time of ~0.18 seconds per image, making it highly efficient for real-time deployment. Validation against a curated dataset of 1.3 GB, containing high-resolution karyotype images, demonstrates superior recall (96.5%) and precision (96.2%), achieving an F1-score of 96.4%, setting new benchmarks in the field.