Chromosome Image Classification Using Edge Fusion Attention Network.

Area of Science:

• Genetics
• Computational Biology
• Medical Imaging

Background:

• Accurate chromosome identification is crucial for karyotype generation and predicting genetic diseases.
• Traditional methods struggle with chromosomal structural variations and boundary detection.

Purpose of the Study:

• To introduce the Edge Fusion Attention Network (EFANet), a novel deep learning architecture for enhanced chromosome classification.
• To overcome limitations of conventional methods in identifying chromosomal abnormalities.

Main Methods:

• Developed EFANet, integrating Adaptive Edge Preserve Fusion (AEPF) for boundary identification and Feature Focused Attention Network (F²ANet) for feature extraction and classification.
• AEPF combines edge and intensity features to highlight morphological differences.
• F²ANet incorporates feature extraction, channel/spatial attention, and classification blocks.

Main Results:

• EFANet achieved high performance: 99.5% accuracy, 99.48% F1 score, 99.63% precision, and 99.45% recall.
• The model demonstrated superior edge detection capabilities, crucial for identifying abnormalities.
• Significantly improved automated chromosome analysis and karyotyping.

Conclusions:

• EFANet offers a robust solution for accurate chromosome classification, surpassing traditional methods.
• The enhanced identification of chromosomal features and abnormalities leads to more precise genetic disorder diagnosis.
• Improved diagnostic accuracy promises better patient outcomes through timely interventions.