染色体核型图多因素量化评分模型的构建^*

Construction of a Multi-Factor Quantitative Scoring Model for Chromosome Karyotype Analysis

摘要

目的 设计多因素量化染色体核型图像质量评价模型用于辅助人工选图，提高选图的效率和染色体核型分析的正确率。方法 通过已有研究和临床专家意见选择染色体交叉度、长度、清晰度、分散度、弯曲度等5项指标作为构建图片质量评价模型的量化指标，设计相应的算法对其进行评分。以专家人工审核结果为金标准，建立染色体核型图片分析结果是否正确的数据集。通过单因素分析和多因素分析判断各量化指标在核型识别正确和错误之间是否有差异，确定图片质量评价指标，并构建选图质量指数，建立核型分析结果预测模型。结果 共对2 708例患者的13 540张染色体图片进行了人工分析，经专家组审核，其中416张存在核型识别错误。对每个图片质量候选指标设定测量算法并计算得分，经过单因素和多因素分析，染色体弯曲度、交叉度、清晰度及长度评分均能影响染色体核型识别准确性，而分散度评分在识别正确组和错误组之间差异并无统计意义（P=0.133）；根据所得的影响因素构建图片质量指数，经过检验，质量指数每增加1分，分析结果出错的风险就能降低8.4%。随着质量指数分数的升高，错误率呈现下降趋势，质量指数为75分以下的图片错误率为3.88%，75分以上图片的错误率为1.95%。基于质量指数建立染色体核型结果是否正确的预测模型，AUC = 0.72（95%CI：0.67~0.77，P < 0.001），展示良好的预测效果。结论 图片质量指数和染色体核型评价结果预测模型能够更加有效地辅助人工选图。

Abstract

Objective To design a multi-factor quantitative evaluation model for chromosomal karyotype image quality to assist manual image selection, thereby improving the efficiency of image selection and the accuracy of karyotype analysis. Methods Based on existing literature and clinical expert opinion, five indicators-chromosomal crossover degree, length, clarity, dispersion, and curvature-were selected as quantitative measures for constructing the image quality evaluation model, and corresponding algorithms were designed to score each indicator. Using expert manual review results as the gold standard, a dataset comprising chromosomal karyotype images labeled with whether the analytical results were correct was established. Univariate and multivariate analyses were performed to determine whether significant differences existed in each quantitative indicator between karyotypically correct and incorrect images. Indicators demonstrating significant associations were retained to construct a selection quality index and to establish a predictive model for karyotype analysis outcomes. Results A total of 13 540 chromosomal images from 2 708 patients underwent manual analysis, of which 416 images were identified by the expert panel as containing karyotype recognition errors. Measurement algorithms were developed and scores were calculated for each candidate image quality indicator. Univariate and multivariate analyses revealed that the scores for chromosomal curvature, crossover degree, clarity, and length significantly influenced the accuracy of karyotype recognition, whereas no statistically significant difference in dispersion score was observed between the correct and incorrect recognition groups (P = 0.133). Based on the identified contributing factors, a composite image quality index was constructed. Validation demonstrated that for every 1-point increase in the quality index, the risk of analytical error decreased by 8.4%. A declining trend in error rates was observed with increasing quality index scores; the error rate among images with a quality index below 75 was 3.88%, compared with 1.95% among images scoring 75 or above. A predictive model for the correctness of karyotype analysis outcomes, established based on the quality index, demonstrated favorable predictive performance with an area under the ROC curve (AUC) of 0.72 (95% CI: 0.67-0.77,P < 0.001). Conclusion The image quality index and the predictive model for karyotype evaluation outcomes can effectively assist manual image selection in chromosomal karyotype analysis.

关键词

染色体核型分析 / 图像质量评价 / 人工智能 / 细胞遗传学

Key words

Chromosomal karyotype analysis / Image quality evaluation / Artificial Intelligence / Cytogenetics

参考文献

[1] Liu X, Liu S, Wang H, et al. Potentials and challenges of chromosomal microarray analysis in prenatal diagnosis[J/OL]. Front Genet, 2022, 13: 938183. https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2022.938183/full.

[2] 中华医学会医学遗传学分会临床遗传学组, 中国医师协会医学遗传医师分会遗传病产前诊断专业委员会, 中华预防医学会出生缺陷预防与控制专业委员会遗传病防控学组. 低深度全基因组测序技术在产前诊断中的应用专家共识 [J]. 中华医学遗传学杂志, 2019, 36(4): 293. DOI: 10.3760/cma.j.issn.1003-9406.2019.04.001.

[3] Clinical Genetics Group, Medical Genetics Branch, Chinese Medical Association, Professional Committee for Prenatal Diagnosis of Genetic Diseases, Medical Genetics Branch, Chinese Medical Association, Group of Genetics Disease Prevention and Control, Birth Defect Prevention and Control Committee, Chinese Society of Preventive Medicine. Expert consensus on the application of low-depth whole genome sequencing in prenatal diagnosis[J]. Chinese Journal of Medical Genetics, 2019, 36(4): 293. DOI: 10.3760/cma.j.issn.1003-9406.2019.04.001.

[4] 罗纯龙, 赵屹. 染色体核型分析深度学习方法综述 [J]. 中国图象图形学报, 2023, 28(11): 3363. DOI: 10.11834/jig.221094.

[5] Luo C L, Zhao Y. Review of deep learning methods for karyotype analysis [J]. Journal of Image and Graphics, 2023, 28(11): 3363. DOI: 10.11834/jig.221094.

[6] Song G, Su L, Wang X, et al. Deep learning-based chromosome segmentation and extraction: a comprehensive review of methodologies, challenges, and future directions[J/OL]. Neurocomputing, 2025, 652: 131060. https://doi.org/10.1016/j.neucom.2025.131060.

[7] Wang C, Yu L, Su J, et al. Fully automatic karyotyping via deep convolutional neural networks [J]. IEEE Access, 2024, 12: 46081. DOI: 10.1109/ACCESS.2024.3380829.

[8] Dinu A G, Biju V G, Vinod B R, et al. Chromenhance-x: a multi-modal deep learning framework for chromosome image enhancement [C/OL] // 2024 3rd Edition of IEEE Delhi Section Flagship Conference, 2024: 1-7. https://esource.scu.edu.cn/https/443/org/ieee/ieeexplore/yitlink/document/10867202.

[9] Wang G, Liu H, Yi X, et al. ARMS Net: overlapping chromosome segmentation based on adaptive receptive field multi-scale network[J/OL]. Biomed Signal Process Control, 2021, 68: 102811. https://doi.org/10.1016/j.bspc.2021.102811.

[10] Moattari M. Precise classification of low quality G-banded chromosome Images by reliability metrics and data pruning classifier[PP/OL]. 2025. https://www.semanticscholar.org/paper/Precise-classification-of-low-quality-G-banded-by-Moattari/5916fa3c34b99da74bbbbee7950698f7d13946e1. DOI: 10.48550/arXiv.2510.21827.

[11] Song S, Wang J, Cheng F, et al. A robust framework of chromosome straightening with ViT-Patch GAN[C/OL] // 2023 IEEE 20th International Symposium on Biomedical Imaging(ISBI). 2023: 1-5. https://esource.scu.edu.cn/https/443/org/ieee/ieeexplore/yitlink/document/10230388/. DOI: 10.1109/ISBI53787.2023.10230388.

[12] Tseng J J, Lu C H, Li J Z, et al. An open dataset of annotated metaphase cell images for chromosome identification[J/OL]. Sci Data, 2023, 10(1): 104. https://www.nature.com/articles/s41597-023-02003-7. DOI: 10.1038/s41597-023-02003-7.

[13] Liu X, Fu L, Chun‐Wei Lin J, et al. SRAS‐net: low‐resolution chromosome image classification based on deep learning[J]. IET Syst Biol, 2022, 16(3/4): 85. DOI: 10.1049/syb2.12042.

[14] 高效智能染色体核型精准辅助诊断系统的临床应用协作组, 王昊, 文娟, 等. 高效智能染色体核型精准辅助诊断系统的临床应用专家共识 [J]. 中华医学遗传学杂志, 2024, 41(9): 1025. DOI: 10.3760/cma.j.cn511374-20240706-00375.

[15] Efficient Intelligent Chromosome Karyotyping Precise Auxiliary Diagnosis System Clinical Application Collaboration Group, Wang H, Wen J, et al. Expert consensus on the clinical application of efficient intelligent chromosomal karyotyping precise auxiliary diagnosis system[J]. Chinese Journal of Medical Genetics, 2024, 41(9): 1025. DOI: 10.3760/cma.j.cn511374-20240706-00375.

[16] Arsham M S, Barch M J, Lawce H J. The AGT cytogenetics laboratory manual[M]. 4th ed. New Jersey: Wiley-Blackwell & Hoboken, 2017: 67-68.

[17] Sharma M, Saha O, Sriraman A, et al. Crowdsourcing for chromosome segmentation and deep classification [C/OL] // 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). 2017: 786-793. https://esource.scu.edu.cn/https/443/org/ieee/ieeexplore/yitlink/document/8014843. DOI: 10.1109/CVPRW.2017.109.

基金资助

*国家重点研发计划(2023YFC2705700)