An automatic progressive chromosome segmentation approach using deep learning with traditional image processing.

Area of Science:

• Genetics and Bioinformatics
• Medical Imaging Analysis
• Computational Biology

Background:

• Accurate chromosome segmentation is vital for genetic disease detection and reducing cytogenetic expert workload.
• Challenges in automatic segmentation include non-rigid chromosome structures and unpredictable clusters of touching or overlapping chromosomes.
• Existing methods struggle with the complexity of metaphase chromosome images.

Purpose of the Study:

• To develop a fully automatic progressive segmentation approach for metaphase chromosome images.
• To improve the accuracy and efficiency of chromosome segmentation in genetic analysis.
• To address the challenge of segmenting overlapping and touching chromosomes.

Main Methods:

• A three-stage progressive segmentation approach combining deep learning and traditional image processing.
• Stage 1: Thresholding-based and geometric methods for initial chromosome and cluster separation.
• Stage 2: A novel chromosome cluster identification network (CCI-Net) for classifying cluster types.
• Stage 3: Integration of traditional image processing with deep Convolutional Neural Networks (CNNs) for instance segmentation of chromosomes from identified clusters.

Main Results:

• The proposed method achieved 94.60% accuracy in chromosome cluster identification.
• Instance segmentation accuracy reached 99.15% on a clinical dataset of 1148 metaphase chromosome images.
• The approach effectively handles complex chromosome clusters and performs fully automatic segmentation.

Conclusions:

• The automatic progressive segmentation method significantly enhances the accuracy of chromosome analysis.
• The CCI-Net and integrated segmentation strategy successfully address challenges posed by chromosome clusters.
• This automated system offers a valuable tool for genetic disease detection and cytogenetic analysis.