Chromosome Image Classification Using Edge Fusion Attention Network

Identification of chromosome pairs is important for karyotype generation and genetic disease prediction. Conventional methods tend to fail with structural variations of the chromosomes and imprecise boundaries. To solve these issues, we propose a deep learning architecture, Edge Fusion Attention Network (EFANet), for chromosome classification. It is a new architecture that combines the Adaptive Edge Preserve Fusion (AEPF) algorithm with the Feature Focused Attention Network (F²ANet). The AEPF algorithm clearly identifies chromosome boundaries and highlights the morphological differences. It improves feature representation by combining edge features with intensity, thereby ensuring accurate classification. The F²ANet block in EFANet improves classification using three main components: a feature extraction block, an attention block with both channel and spatial attention, and a classification block. Our proposed method ensures accurate chromosome classification, which is essential for diagnosing genetic disorders such as aneuploidies and translocations. Edge detection, a key feature of EFANet, enhances the identification of chromosome abnormalities by focusing more on unusual shapes than normal ones. Our proposed EFANet showed strong performance with 99.5% accuracy, 99.48% F1 score, 99.63% precision, and 99.45% recall. These results highlight its effectiveness in edge detection, which is important for improving automated chromosome analysis. This approach tremendously improves karyotyping by overcoming the limitations of the traditional approaches, resulting in more accurate and timely genetic disease diagnosis and eventually better patient outcomes.