Development of an Artificial Intelligence-Based Chromosome Interpretation System for Amniotic Fluid Karyotyping

Area of Science:

• Cytogenetics
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Conventional G-banded karyotyping is crucial for prenatal diagnosis but is labor-intensive and requires expert interpretation.
• Automating chromosome analysis can address the challenges of manual karyotyping.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) workflow for automated chromosome interpretation from amniotic fluid metaphase images.
• To assess the accuracy and efficiency of the AI system in supporting cytogenetic laboratories.

Main Methods:

• Developed a modular AI workflow including image denoising, chromosome segmentation, overlap screening, and morphology-based classification.
• Trained the system on 13,223 clinical cases with over 50,000 manually annotated chromosomes.
• Validated the workflow on independent testing cohorts and applied it to unsorted clinical metaphase images.

Main Results:

• The AI workflow achieved high classification accuracy across training (97.45%), validation (96.95%), and testing (95.72%) cohorts.
• An overlap-recognition module effectively identified composite chromosome regions, reducing downstream errors.
• The system generated draft karyotypes from unsorted images with high concordance to expert review.

Conclusions:

• An AI-assisted pipeline can significantly streamline labor-intensive karyotyping steps.
• The developed workflow enhances diagnostic efficiency in cytogenetic laboratories while maintaining interpretive reliability.
• AI holds potential to revolutionize prenatal diagnosis through automated chromosome analysis.