Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Chromosomal Structural Abnormality Diagnosis by Homologous Similarity

Juren Li, Fanzhe Fu, Ran Wei, Yifei Sun, Zeyu Lai, Ning Song, Xin Chen, Yang Yang

TL;DR

This work tackles the challenging task of diagnosing chromosomal structural abnormalities, which are rare, morphologically diverse, and distribution-variant across hospitals. It introduces HomNet, a homologous-similarity driven framework consisting of CMSBlock, HomBlock, and BagBlock, trained via self-supervised learning with artificial abnormalities and then fine-tuned for each hospital. Across four real-world hospital datasets and a public dataset with artificial abnormalities, HomNet consistently outperforms baselines, with ablation studies confirming the value of pairwise alignment, multi-pair aggregation, and regional feature modeling. The approach is integrated into AutoVision for fast, millisecond-scale clinical inference, achieving high diagnostic accuracy and enabling scalable, multi-center chromosomal analysis in practice.

Abstract

Pathogenic chromosome abnormalities are very common among the general population. While numerical chromosome abnormalities can be quickly and precisely detected, structural chromosome abnormalities are far more complex and typically require considerable efforts by human experts for identification. This paper focuses on investigating the modeling of chromosome features and the identification of chromosomes with structural abnormalities. Most existing data-driven methods concentrate on a single chromosome and consider each chromosome independently, overlooking the crucial aspect of homologous chromosomes. In normal cases, homologous chromosomes share identical structures, with the exception that one of them is abnormal. Therefore, we propose an adaptive method to align homologous chromosomes and diagnose structural abnormalities through homologous similarity. Inspired by the process of human expert diagnosis, we incorporate information from multiple pairs of homologous chromosomes simultaneously, aiming to reduce noise disturbance and improve prediction performance. Extensive experiments on real-world datasets validate the effectiveness of our model compared to baselines.

Chromosomal Structural Abnormality Diagnosis by Homologous Similarity

TL;DR

This work tackles the challenging task of diagnosing chromosomal structural abnormalities, which are rare, morphologically diverse, and distribution-variant across hospitals. It introduces HomNet, a homologous-similarity driven framework consisting of CMSBlock, HomBlock, and BagBlock, trained via self-supervised learning with artificial abnormalities and then fine-tuned for each hospital. Across four real-world hospital datasets and a public dataset with artificial abnormalities, HomNet consistently outperforms baselines, with ablation studies confirming the value of pairwise alignment, multi-pair aggregation, and regional feature modeling. The approach is integrated into AutoVision for fast, millisecond-scale clinical inference, achieving high diagnostic accuracy and enabling scalable, multi-center chromosomal analysis in practice.

Abstract

Pathogenic chromosome abnormalities are very common among the general population. While numerical chromosome abnormalities can be quickly and precisely detected, structural chromosome abnormalities are far more complex and typically require considerable efforts by human experts for identification. This paper focuses on investigating the modeling of chromosome features and the identification of chromosomes with structural abnormalities. Most existing data-driven methods concentrate on a single chromosome and consider each chromosome independently, overlooking the crucial aspect of homologous chromosomes. In normal cases, homologous chromosomes share identical structures, with the exception that one of them is abnormal. Therefore, we propose an adaptive method to align homologous chromosomes and diagnose structural abnormalities through homologous similarity. Inspired by the process of human expert diagnosis, we incorporate information from multiple pairs of homologous chromosomes simultaneously, aiming to reduce noise disturbance and improve prediction performance. Extensive experiments on real-world datasets validate the effectiveness of our model compared to baselines.
Paper Structure (21 sections, 11 equations, 4 figures, 5 tables)

This paper contains 21 sections, 11 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Dataset and Preprocessing
  4. Problem definition
  5. Methodology
  6. Overview
  7. Self-supervised Learning with Artificial Abnormalities
  8. Modeling the Feature of Chromosomes
  9. Capture the Differences between Homologous Chromosomes
  10. Experiment
  11. Experiment Setting
  12. Performance Evaluation
  13. Ablation Study
  14. Running Time Comparison
  15. Application
  16. ...and 6 more sections

Figures (4)

  • Figure 1: The homologous chromosomes with structural abnormality. (a) shows corresponding healthy region where structural abnormality occurred on (b). As for (c) and (d) are the data preprocessing we applied. (e) and (f) are the embedding of (c) and (d) by CMSBlock from HomNet model.
  • Figure 2: The chromosomal sequences are the gray mean sequences of the left and right parts of the chromosome diagram.
  • Figure 3: The framework of HomNet. The arrow heads refer to the dataflow of a pair of homologous chromosomes.
  • Figure 4: AutoVision, equipped with HomNet, is an advanced intelligent chromosome karyotype analysis system, which registered in NMPA (National Medical Products Administration)and obtained CE mark.