Scholarlink

<p>Automated classification of thyroid Fine Needle Aspiration Biopsy (FNAB) images faces challenges in limited data, inter-observer variability, and computational cost. Efficient, interpretable models are crucial for clinical support.&#xa0;We developed and externally validated a deep learning system for clinically-oriented tri-class classification of thyroid FNAB images. This system classifies images into three key categories (Bethesda II: Observation / Non– Surgical Management (Benign), Bethesda I, III, IV, V: Further Investigation / Consideration for Intervention (Indeterminate/Suspicious), and Bethesda VI: Surgical Intervention (Malignant)) that correspond to post-biopsy treatment pathways in Vietnam, a developing country. Our goal was to achieve high diagnostic accuracy with low computational overhead to potentially serve as a decision-support tool.&#xa0;Our pipeline features: (1) YOLOv10-based cell cluster detection for informative sub-region extraction and noise reduction; (2) a curriculum learning-inspired protocol sequencing localized crops to full images for multi-scale feature capture; (3) adaptive lightweight EfficientNetB0 (4 million parameters) selection balancing performance and efficiency; and (4) a Transformer-inspired module for multi-scale, multi-view feature integration. External validation used 1,015 independent FNAB images.&#xa0;ThyroidEffi Basic achieved a macro F1 of 89.19% and Area Under the Curve (AUC)s of 0.98 (Benign), 0.95 (Indeterminate/Suspicious), and 0.96 (Malignant) on the internal test set. External validation yielded AUCs of 0.9495 (Benign), 0.7436 (Indeterminate/Suspicious), and 0.8396 (Malignant). ThyroidEffi Premium improved macro F1 to 89.77%. Gradient-weighted Class Activation Mapping (Grad-CAM) highlighted key diagnostic regions, supporting interpretability. The system processed 1000 cases in 55 seconds, highlighting the challenges of cross-site domain shift in resource-constrained settings.&#xa0;This study suggests that high-accuracy, interpretable thyroid FNAB image classification is achievable with minimal computational demands.</p> Graphical Abstract <p>Figure provides a concise visual summary of our proposed methodology and its key contributions.</p>

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

ThyroidEffi 1.0: a cost-effective system for high-performance multi-class thyroid carcinoma classification

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Hai Pham Ngoc,
  • De Nguyen Van,
  • Dung Vu Tien,
  • Phuong Le-Hong

摘要

Automated classification of thyroid Fine Needle Aspiration Biopsy (FNAB) images faces challenges in limited data, inter-observer variability, and computational cost. Efficient, interpretable models are crucial for clinical support. We developed and externally validated a deep learning system for clinically-oriented tri-class classification of thyroid FNAB images. This system classifies images into three key categories (Bethesda II: Observation / Non– Surgical Management (Benign), Bethesda I, III, IV, V: Further Investigation / Consideration for Intervention (Indeterminate/Suspicious), and Bethesda VI: Surgical Intervention (Malignant)) that correspond to post-biopsy treatment pathways in Vietnam, a developing country. Our goal was to achieve high diagnostic accuracy with low computational overhead to potentially serve as a decision-support tool. Our pipeline features: (1) YOLOv10-based cell cluster detection for informative sub-region extraction and noise reduction; (2) a curriculum learning-inspired protocol sequencing localized crops to full images for multi-scale feature capture; (3) adaptive lightweight EfficientNetB0 (4 million parameters) selection balancing performance and efficiency; and (4) a Transformer-inspired module for multi-scale, multi-view feature integration. External validation used 1,015 independent FNAB images. ThyroidEffi Basic achieved a macro F1 of 89.19% and Area Under the Curve (AUC)s of 0.98 (Benign), 0.95 (Indeterminate/Suspicious), and 0.96 (Malignant) on the internal test set. External validation yielded AUCs of 0.9495 (Benign), 0.7436 (Indeterminate/Suspicious), and 0.8396 (Malignant). ThyroidEffi Premium improved macro F1 to 89.77%. Gradient-weighted Class Activation Mapping (Grad-CAM) highlighted key diagnostic regions, supporting interpretability. The system processed 1000 cases in 55 seconds, highlighting the challenges of cross-site domain shift in resource-constrained settings. This study suggests that high-accuracy, interpretable thyroid FNAB image classification is achievable with minimal computational demands.

Graphical Abstract

Figure provides a concise visual summary of our proposed methodology and its key contributions.