Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Accelerating Lung Cancer Diagnosis: Label-Free Pathology Through Deep Learning
Researchers have developed a label-free method using autofluorescence imaging and deep learning to diagnose non-small cell lung cancer subtypes without conventional tissue staining, achieving expert-validated accuracy while reducing diagnostic time and costs.
Background
Accurate subtyping of non-small cell lung cancer (NSCLC)—distinguishing adenocarcinoma from squamous cell carcinoma—is critical for treatment selection. Current diagnostic workflows rely on immunohistochemical staining with markers like TTF-1 and p40, adding time and expense.
Key Findings
The study achieved exceptional classification performance:
- Area under curve (AUC) of 0.981 for binary classification of NSCLC subtypes
- AUC of 0.996 for multi-class classification distinguishing non-cancerous tissue and multiple cancer subtypes
- Virtual immunohistochemical stains (TTF-1 and p40) validated by three experienced thoracic pathologists
The method analyzed unstained tissue samples using autofluorescence imaging (intensity and fluorescence lifetime) combined with convolutional neural networks and generative adversarial networks for classification and virtual stain generation.
Why It Matters
This label-free approach significantly accelerates diagnostic workflows while eliminating conventional staining costs. Clinical-grade virtual IHC enables rapid, confident pathologist decisions without sacrificing accuracy—essential for timely treatment planning.
Limitations
The study focused specifically on NSCLC subtyping and did not explore broader applicability beyond this indication.
Original paper: Label-free pathological subtyping of non-small cell lung cancer using deep classification and virtual immunohistochemical staining. — NPJ digital medicine. 10.1038/s41746-026-00356-1
