Edge Preserved U-Net for chromosome edge detection

Area of Science:

• Medical Imaging
• Computational Biology
• Deep Learning

Background:

• Accurate chromosome boundary detection is crucial for karyotype analysis, enabling precise classification and arrangement.
• Existing segmentation techniques often struggle with preserving fine boundary details in complex chromosome structures.

Purpose of the Study:

• To introduce an Edge Preserved U-Net (EPU-Net) architecture for robust chromosome boundary detection.
• To enhance the accuracy and efficiency of karyotype analysis through improved segmentation.

Main Methods:

• The proposed EPU-Net integrates a guided filter and Sobel edge detector within a U-Net architecture.
• Incorporation of Edge Preserving Skip Layers (EPSL) and Edge Preserving Blocks (EPB) with dilated convolutions to maintain spatial information and extract multi-scale features.
• The architecture includes preprocessing blocks, convolution, pooling, and upsampling operations with skip connections.

Main Results:

• EPU-Net achieved superior segmentation accuracy (99.80%) and Intersection over Union (IoU) score (99.60%) on a custom dataset.
• The model demonstrated high recall (99.89%) and specificity (99.71%), indicating robust performance.
• A Structure Similarity Index Metric (SSIM) of 0.9985 highlights the preservation of structural integrity.

Conclusions:

• The EPU-Net model offers a significant advancement in chromosome boundary detection for karyotype analysis.
• The proposed deep learning architecture demonstrates robustness and outperforms existing state-of-the-art segmentation methods.
• This approach facilitates more accurate and reliable automated karyotyping.