Implementing trust in non-small cell lung cancer diagnosis with a conformalized uncertainty-aware AI framework

2026-06-23 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Medical Devices & Health Technology, Health & Medical Research, Clinical Care & Medical Practice · Depth: Expert, extended

Summary

TRUECAM (Trustworthiness-focused, Uncertainty-aware, End-to-end Cancer diagnosis with Model-agnostic capabilities) is a novel AI framework for non-small cell lung cancer (NSCLC) subtyping using whole-slide images. It integrates a spectral-normalized neural Gaussian process (SNGP) for out-of-scope input identification, ambiguity-guided tile elimination (EAT) to filter ambiguous regions, and conformal prediction (CP) for controlled error rates. Evaluated across multiple cancer datasets, including over 20,000 WSIs, and with both task-specific (Inception-v3) and foundation models (UNI, CONCH, Prov-GigaPath, TITAN), TRUECAM consistently improved classification accuracy, robustness, interpretability, data efficiency, and fairness. For instance, Inception-v3 with TRUECAM reduced error rates by 72.0% (1-α=0.95) and 93.8% (1-α=0.99). It also reduced racial accuracy gaps by 38.1% (TCGA) and 78.3% (CPTAC).

Key takeaway

For AI Scientists developing diagnostic tools, integrating TRUECAM into your pathology AI pipeline is crucial for deploying trustworthy systems. This framework allows you to achieve statistically guaranteed error rates, robustly handle out-of-domain data, and enhance fairness across demographic groups, significantly reducing misdiagnosis risks. Consider adopting its SNGP, EAT, and CP components to improve model reliability and interpretability in real-world clinical settings.

Key insights

TRUECAM enhances pathology AI trustworthiness by integrating uncertainty quantification, data filtering, and error rate control.

Principles

• Data trustworthiness requires identifying out-of-scope inputs.
• Model trustworthiness needs statistically guaranteed error rates.
• Eliminating ambiguous data improves model reliability.

Method

TRUECAM combines SNGP for OOD detection and uncertainty estimation, ambiguity-guided tile elimination (EAT) to filter noisy regions, and conformal prediction (CP) with conformal risk control (CRC) for statistically guaranteed error rates.

In practice

• Use SNGP to detect out-of-domain whole-slide images.
• Apply EAT to remove ambiguous tiles, reducing computational load.
• Implement CP to ensure a 1-α coverage for predictions.

Topics

• Pathology AI
• Non-small Cell Lung Cancer
• Uncertainty Quantification
• Conformal Prediction
• Whole-Slide Imaging
• AI Trustworthiness

Code references

• iamownt/TRUECAM

Best for: AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• Look-Closer-Then-Diagnose: Confidence-Aware Ultrasound VQA via Active Zooming — A novel VLM framework integrates lesion-focused reasoning and uncertainty-aware rewards for improved ultrasound diagnosis.
• Dual-Modal Lung Cancer AI: Interpretable Radiology and Microscopy with Clinical Risk Integration — Integrating CT radiology and H&E microscopy with clinical data significantly enhances lung cancer diagnosis and interpretability.
• From Black-Box Confidence to Measurable Trust in Clinical AI: A Framework for Evidence, Supervision, and Staged Autonomy — Clinical AI trust is an engineered system property, not just model accuracy, requiring evidence, supervision, and staged autonomy.
• Margin-Consistent Deep Subtyping of Invasive Lung Adenocarcinoma via Perturbation Fidelity in Whole-Slide Image Analysis — Margin consistency with Perturbation Fidelity enhances deep learning robustness and accuracy for lung adenocarcinoma subtyping.
• Trusting Right Predictions for Wrong Reasons: A LIME Based Analysis of Deep Learning Interpretability in Lung Cancer Diagnosis — Deep learning models can achieve high predictive accuracy in medical imaging for different reasons, requiring independent interpretability validation.
• Case-Aware Medical Image Classification with Multimodal Knowledge Graphs and Reliability-Guided Refinement — Medical image classification improves by integrating multimodal knowledge graphs from similar cases with reliability-guided refinement.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.