We also have X and podcasts
AI Identifies Tumor Origins from Tissue Images with 98% Accuracy
A new deep learning method identifies where malignant tumors originated by analyzing whole slide images with 98% accuracy, processing each image in under a minute.
Background
Cancers of unknown primary (CUP) are difficult to diagnose when the tumor’s origin is unknown. Traditional methods require immunohistochemical staining and extensive investigation. The ETMIL-SL-SMFA model offers a faster alternative, analyzing digital tissue images using transformer-based machine learning.
Key Findings
- 98±1% accuracy and 99±1% AUROC on histopathology datasets; 97±1% for breast cancers, 96±1% for lung cancers
- 86% accuracy on cytological smears, outperforming four existing methods
- Processes whole slide images in 11–36 seconds with 99%+ space efficiency
- Trains in under 10 hours on a single TITAN RTX GPU
Why It Matters
This enables rapid diagnosis of CUP without immunohistochemical testing. Its computational efficiency makes it practical for real-world clinical deployment, including resource-limited settings where advanced diagnostics are most needed.
Limitations
Sample sizes for some datasets were modest, and accuracy varied across cancer types. Further validation in independent clinical cohorts will be necessary before widespread implementation.
Original paper: Soft multiclass feature augmented deep learning to predict tumor origins using cytology or histology whole slide images. — NPJ digital medicine. 10.1038/s41746-026-02604-7
