Scholarlink

<p>The early and accurate diagnosis of head and neck tumors remains a clinical challenge due to the diversity of anatomical sites, histopathological subtypes, and imaging modalities. Artificial intelligence (AI) has emerged as a promising tool for augmenting tumor detection and classification, but the evidence supporting its diagnostic accuracy varies considerably across studies. This systematic review and meta-analysis synthesized evidence from 6 published meta-analyses and 14 original studies, capturing data up to November 2025. Studies assessing AI-based models (machine learning &amp; deep learning) for the diagnosis of head and neck tumors were included, analyzing their pooled sensitivity, specificity, area under the curve (AUC), heterogeneity, and risk of bias (QUADAS-2). Subgroup analyses were performed by imaging modality and AI method. A total of 110 datasets (including 96 from meta-analyses and 14 high-quality individual studies) were analyzed. The overall pooled diagnostic sensitivity was 87.5% (95% CI: 0.842–0.907), specificity was 87.5% (95% CI: 0.846–0.903), with an AUC of 0.91. CNN-based deep learning models demonstrated the highest accuracy (AUC: 0.94), particularly when applied to histopathology and CT imaging. Significant heterogeneity (I² 60–85%) reflected the variety of tumor sites, AI architectures, imaging modalities, and study populations. The QUADAS-2 assessment indicated robust methodological quality for most studies. AI models, particularly deep learning, provide robust and accurate diagnostic support for head and neck tumors, demonstrating performance comparable to or exceeding human experts in certain tasks. Prospective, standardized, and multi-institutional studies are needed to confirm real-world clinical utility and ensure broad applicability.</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

Diagnostic Performance of Artificial Intelligence Models for Head and Neck Tumours

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Kartavya Kumar Verma

摘要

The early and accurate diagnosis of head and neck tumors remains a clinical challenge due to the diversity of anatomical sites, histopathological subtypes, and imaging modalities. Artificial intelligence (AI) has emerged as a promising tool for augmenting tumor detection and classification, but the evidence supporting its diagnostic accuracy varies considerably across studies. This systematic review and meta-analysis synthesized evidence from 6 published meta-analyses and 14 original studies, capturing data up to November 2025. Studies assessing AI-based models (machine learning & deep learning) for the diagnosis of head and neck tumors were included, analyzing their pooled sensitivity, specificity, area under the curve (AUC), heterogeneity, and risk of bias (QUADAS-2). Subgroup analyses were performed by imaging modality and AI method. A total of 110 datasets (including 96 from meta-analyses and 14 high-quality individual studies) were analyzed. The overall pooled diagnostic sensitivity was 87.5% (95% CI: 0.842–0.907), specificity was 87.5% (95% CI: 0.846–0.903), with an AUC of 0.91. CNN-based deep learning models demonstrated the highest accuracy (AUC: 0.94), particularly when applied to histopathology and CT imaging. Significant heterogeneity (I² 60–85%) reflected the variety of tumor sites, AI architectures, imaging modalities, and study populations. The QUADAS-2 assessment indicated robust methodological quality for most studies. AI models, particularly deep learning, provide robust and accurate diagnostic support for head and neck tumors, demonstrating performance comparable to or exceeding human experts in certain tasks. Prospective, standardized, and multi-institutional studies are needed to confirm real-world clinical utility and ensure broad applicability.