报告详情
Design of an Integrated Quality-Adaptive Multi-Scale Causal Ensemble Framework for Explainable and Risk-Calibrated Screening of Oral Potentially Malignant Disorders
编号：70
访问权限：仅限参会人
更新：2026-11-10 16:36:46 浏览：7次
Online
报告开始：2026年11月21日 14:00（Asia/Kolkata）
报告时间：15min
所在会场：[S4] Artificial Intelligence and Intelligent Computing [S4-8] Artificial Intelligence and Intelligent Computing
暂无文件
摘要
There's an urgent need for intelligent screening of oral potentially malignant disorders (OPMDs) so that there can be better early referral to specialists and fewer diagnostic delays. Most current machine learning models have some limitations due to variable lighting conditions; poor or weak localization of lesions; use of a fixed scale representation; the use of constant weighting factors when combining multiple predictors; reliance on shortcuts in explaining how they make their predictions; and poorly calibrated decision-making across centers. In this paper, we propose the Q-MorphoFuse five stage analytic methodology. Q-LESRNet will perform quality conditioned restoration, evidence-based segmentation, and boundary uncertainty estimation. The MorphoSpectral Pyramid Tokenizer will extract features at each level of granularity (spatial, morphology, spectral frequency) which preserves both local subtle textures as well as global architectural characteristics around lesions. CRED Fuse will then combine these feature vectors using the results of three different metrics - causal reliability, epistemic uncertainty, and expert disagreement to produce patient specific ensemble weights. Finally, CoStaX-OPMD will validate model performance against gold standard pathology sensitive and nuisance invariant counterfactual testing thus allowing for identification of potential spurious attentions. Lastly, ConformaRisk-Triage will provide conformal prediction intervals as well as calibrated malignancy risks, and whether it is safe to abstain from further testing or to refer immediately for the process. Overall, our proposed method should lead to improved sensitivity, calibration, explainability, device portability/robustness, and safety in identifying high risk OPMD patients while also decreasing the number of false negatives and unnecessary referrals in resource challenged clinical environments.
关键词
Oral Potentially Malignant Disorders, Intraoral Image Analysis, Multi-Scale Deep Learning, Causal Ensemble Fusion, Conformal Risk Prediction, Scenarios
报告人
KIRAN KUMAR CHANUMOLU
ASSISTANT PROFESSOR K L University稿件作者
全部评论
重要日期
-
会议日期
11月20日
2026
至11月22日
2026
-
10月31日 2026
初稿截稿日期
-
11月07日 2026
提前注册日期
-
11月22日 2026
注册截止日期
主办单位
United Societies of Science
承办单位
Pondicherry University
协办单位
IEEE Section
IEEE Madras Section
IEEE Madras Section
联系方式
- ICon INDIA 2026
- ic******@usssociety.org
发表评论