Scholarlink

Automatic diagnosis of chromosomal anomalies from cytogenetic images is critical for geneticists in detecting abnormalities and making diagnostic decisions. While Convolutional Neural Networks (CNNs) have shown significant potential in detecting genetic abnormalities in chromosomes, but their application faces notable challenges, including limited dataset availability and the inherent lack of interpretability in their decision-making processes.This paper introduces a novel model trained on a unique dataset of 7680 annotated karyograms, specifically curated to improve chromosomal aberration detection. By combining transfer learning using the pre-trained VGG16 model in conjunction with Grad-CAM, the system not only classifies karyograms but also identifies Regions of Interest (ROI), particularly focusing on chromosomal anomalies, and provides visual explanations for its predictions, thus addressing the interpretability gap. This work is the first to apply Explainable AI (XAI) techniques, specifically Grad-CAM, to explain results for chromosomal abnormality detection in karyograms. The model achieves 98% accuracy in detecting chromosome 20 deletion (del(20q)), a key anomaly associated with conditions like Myelodysplastic Syndromes (MDS) and Acute Myeloid Leukemia (AML). This explainable, automated method enhances diagnostic reliability by validating predictions against expert-annotated ROIs, facilitating more accurate and interpretable decisions for cytogeneticists.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Explainable AI Framework for Chromosomal Anomaly Detection in Karyograms Using Grad-CAM and Advanced Neural Networks

  • 章节

  • 发表日期：

    2026年

查看全文
  • Sumaira Tabassum,
  • M. Jawad Khan,
  • Javaid Iqbal,
  • Asim Waris,
  • M. Adeel Ijaz

摘要

Automatic diagnosis of chromosomal anomalies from cytogenetic images is critical for geneticists in detecting abnormalities and making diagnostic decisions. While Convolutional Neural Networks (CNNs) have shown significant potential in detecting genetic abnormalities in chromosomes, but their application faces notable challenges, including limited dataset availability and the inherent lack of interpretability in their decision-making processes.This paper introduces a novel model trained on a unique dataset of 7680 annotated karyograms, specifically curated to improve chromosomal aberration detection. By combining transfer learning using the pre-trained VGG16 model in conjunction with Grad-CAM, the system not only classifies karyograms but also identifies Regions of Interest (ROI), particularly focusing on chromosomal anomalies, and provides visual explanations for its predictions, thus addressing the interpretability gap. This work is the first to apply Explainable AI (XAI) techniques, specifically Grad-CAM, to explain results for chromosomal abnormality detection in karyograms. The model achieves 98% accuracy in detecting chromosome 20 deletion (del(20q)), a key anomaly associated with conditions like Myelodysplastic Syndromes (MDS) and Acute Myeloid Leukemia (AML). This explainable, automated method enhances diagnostic reliability by validating predictions against expert-annotated ROIs, facilitating more accurate and interpretable decisions for cytogeneticists.