CaresAI at SMM4H-HeaRD 2026: Predicting TNM Staging

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Health & Wellbeing — Health & Medical Research, Medical Devices & Health Technology, Clinical Care & Medical Practice · Depth: Expert, quick

Summary

CaresAI's study at SMM4H-HeaRD 2026 focused on predicting Tumor, Node, and Metastasis (TNM) cancer stage labels independently using Cancer Genome Atlas (TCGA) pathology reports. The research framed this as three multi-label classification tasks, exploring both classical and deep learning methods. Features included Term Frequency-Inverse Document Frequency (TF-IDF) and embeddings from ClinicalBERT, BioBERT, and PubMedBERT, applied with Logistic Regression, Light Gradient Boosting Machine, Feed-Forward Neural Networks, and Wide Residual Networks. LightGBM with TF-IDF achieved the best training AUROC scores (0.9368 for T, 0.9524 for N, 0.8311 for M). Test set Macro-F1 scores were high (e.g., 0.978 for T on test set 1), but a notable performance decline from 0.938 to 0.858 was observed between test sets, indicating generalizability issues, class imbalance sensitivity, and challenges with lengthy clinical documents. The study provides an efficient baseline and reproducible pipeline, but requires further optimization for real-world clinical use.

Key takeaway

For AI Scientists and Machine Learning Engineers developing clinical NLP solutions for cancer staging, be aware that models achieving high training performance may struggle with generalizability on diverse test sets. Your focus should extend beyond initial metrics to robust validation, addressing class imbalance, and optimizing for real-world clinical document complexity before deployment.

Key insights

Combining classical and deep learning models with specific embeddings improves TNM cancer staging prediction, though generalizability remains a challenge.

Principles

• Combining embeddings enhances predictive ability.
• Model generalizability is crucial for clinical utility.
• Class imbalance impacts clinical document processing.

Method

Frame TNM staging as three multi-label classification tasks. Utilize TF-IDF and BERT-based embeddings with LR, LightGBM, FFNN, and WRN models for prediction.

In practice

• Use LightGBM with TF-IDF for high AUROC.
• Combine ClinicalBERT, BioBERT, PubMedBERT embeddings.
• Evaluate models on diverse test sets for generalizability.

Topics

• TNM Staging
• Cancer Prediction
• Clinical NLP
• Multi-label Classification
• LightGBM
• BERT Embeddings
• TCGA Pathology Reports

Best for: NLP Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• From Hazard Functions to Language Space: Cox-Supervised Distillation of Survival Risk into a Large Language Model — Large language models can internalize and predict survival risk from structured clinical data via text-based distillation.
• Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Contrastive learning, especially with expert-guided hard negatives, improves fine-grained pediatric brain tumor classification from WSIs.
• Domain Fine-Tuning FinBERT on Finnish Histopathological Reports: Train-Time Signals and Downstream Correlations — Domain fine-tuning FinBERT on medical text can be predicted by observing embedding geometry changes.
• Multi-modal AI for comprehensive breast cancer prognostication — A multi-modal AI test significantly improves breast cancer prognostication by integrating pathology and clinical data.
• Automated Detection of Dosing Errors in Clinical Trial Narratives: A Multi-Modal Feature Engineering Approach with LightGBM — Combining sparse lexical features with dense semantic embeddings improves clinical dosing error detection.
• Prompt, Plan, Extract: Zero-Shot Agentic LLMs Workflows for Lung Pathology Extraction from Clinical Narratives — Zero-shot agentic LLMs can accurately extract complex lung pathology data from clinical narratives without specific training.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.