A chromosome abnormality detection method, system and device

By using collaborative modeling of ResNet18 and Transformer, combined with triple positional encoding, the problem of insufficient local feature capture capability and global correlation in existing technologies is solved, achieving high-precision detection of chromosomal abnormalities, especially accurate identification of small-scale and position-sensitive abnormalities.

CN122177209APending Publication Date: 2026-06-09FOURTH MILITARY MEDICAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
FOURTH MILITARY MEDICAL UNIVERSITY
Filing Date
2026-03-31
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately extract key morphological details such as chromosome band intensity gradients and fine structural breaks, and cannot effectively distinguish the biological positional significance of the 23 pairs of chromosomes. This results in a high rate of missed detection for small-scale structural abnormalities, low accuracy in detecting position-sensitive abnormalities, and poor interpretability.

Method used

After extracting local features of the kernel image using ResNet18, it is converted into an embedding with positional semantics through a position-aware projection layer. Global correlation is modeled by combining Transformer, and end-to-end detection is achieved through a multi-task classification head. Triple positional encoding is incorporated to enhance feature representation.

Benefits of technology

It significantly reduced the false negative rate of small-scale structural abnormalities, improved the detection accuracy of global abnormalities such as transchromosomal translocations, and enhanced the detection precision and interpretability of location-sensitive abnormalities such as trisomy 21.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122177209A_ABST
    Figure CN122177209A_ABST
Patent Text Reader

Abstract

The application discloses a chromosome abnormality detection method, system and device, and relates to the technical field of medical image analysis. The method comprises the following steps: acquiring a chromosome patch sequence of a karyotype image; adopting a pre-trained convolutional neural network to extract local features of each patch in the chromosome patch sequence, converting the local features into an embedded vector sequence containing chromosome position semantics, inputting the embedded vector sequence into a multi-layer Transformer encoder after triple position coding to perform global correlation modeling, and outputting global features and patch features; performing chromosome position recognition based on the patch features to assist chromosome abnormality binary classification based on the global features, and outputting a probability distribution of the chromosome abnormality. The method significantly improves the missed detection rate of small-range structural abnormalities and the detection accuracy of global abnormalities such as translocation between chromosomes.
Need to check novelty before this filing date? Find Prior Art