Researchers (Akram lab) have developed a new imaging technique combined with AI that can accurately identify major types of non-small cell lung cancer (NSCLC), potentially helping patients receive the right treatment sooner than current methods. Lung cancer is the world's leading cause of cancer-related death, with non-small cell lung cancer (NSCLC) accounting for around 80% of cases. Correctly identifying NSCLC subtypes, particularly adenocarcinoma and squamous cell carcinoma, is essential because treatment decisions and patient outcomes depend on accurate diagnosis.Currently, identifying tumour subtypes usually requires several laboratory steps, including staining biopsy tissue samples with special chemicals that bind to markers of interest. These tests can be slow, expensive, and require highly trained specialists.Capturing natural light signalsIn this study, IRR researchers, working with colleagues at the Royal Infirmary of Edinburgh and the University's School of Engineering, developed a new way to analyse lung cancer tissue without the need for traditional staining.The team captured images of natural light signals, known as autofluorescence, emitted by untreated tissue samples. This included using an imaging technique called FLIM (Fluorescence Lifetime Imaging Microscopy) which measures time taken for the fluorescence signal after the tissue has been excited by light. They then trained AI models to distinguish between non-cancerous tissue and different non-small cell lung cancer subtypes from these images.The researchers also developed a virtual staining system using generative AI, which artificially added traditional stains onto the inputted images, and generated digital versions of them. These were stains for proteins TTF-1 (marker used to identify adenocarcinoma) and p40 (marker used to identify squamous cell carcinoma).Non-small cell lung cancer subtypes correctly identifiedThe autofluorescence-based method accurately distinguished between major NSCLC subtypes and non-cancerous tissue.The virtual staining system produced high-quality AI-generated images that closely matched conventional laboratory stains. Independent assessments by experienced thoracic pathologists confirmed that the virtual images were reliable and suitable for diagnostic use. Patient tissue sample showing TTF-1 (brown stain), a marker of lung adenocarcinoma. Left: conventional staining. Right: AI-generated virtual staining from a label-free autofluorescence image. Clinicians judged the virtual image to be reliable for diagnosis and clinical decision-making. The findings suggest that autofluorescence signals contain enough biological information to support both automated tumour classification and virtual visualisation of key diagnostic markers.This is the first study to demonstrate FLIM-based label-free virtual staining for TTF-1 and p40, two important biomarkers used in routine NSCLC diagnosis. The classification and virtual staining systems can be used separately or together to improve diagnostic efficiency and support clinical decision-making. Our approach could allow doctors to diagnose lung cancer more quickly and efficiently, while preserving valuable tissue samples and reducing the need for complex laboratory procedures. Ultimately, this could help patients receive the right treatment sooner without compromising diagnostic accuracy. Professor Ahsan Akram IRR Group Leader and the paper’s co-corresponding author What's next?Although the results are promising, further studies are needed before the technology can be used in clinical practice. The researchers will now test the approach in larger and more diverse patient groups, optimise the imaging workflow, and continue improving cancer subtype identification.Ultimately, they envision a quick, label-free pathology workflow capable of generating multiple virtual stains and accurate tumour classifications from a single image, transforming how lung cancer is diagnosed.This work was supported by Wellcome, Cancer Research, Medical Research Council, Engineering and Physical Sciences Research Council (EPSRC), NVIDIA and UKRI.Read the full article in npj Digital MedicineAkram research group Tags CIR This article was published on Thursday 11 June 2026
