Summary: UCLA researchers in the Departments of Electrical and Computer Engineering & Bioengineering have developed a non-invasive, biopsy-free technique for in vivo virtual histology of skin cancer using deep learning.
Background: Tissue biopsies have often been used to diagnose skin cancers in patients. These procedures are generally cumbersome and invasive, requiring dissection of the affected tissue. The dissected tissue sample is then stained for histological pathology for diagnosis. However, this method often requires one day to several weeks for final diagnosis. Recent advances in emerging optical technologies, such as the reflectance confocal microscopy (RCM) are capable of non-invasive, cellular-level resolution of in vivo images of skin without biopsy. However, this technique requires highly specialized training to correlate the greyscale images with pathological features and the images provided generally lack detailed cellular features. Therefore, there is significant and pressing need to improve upon current non-invasive optical techniques, such as RCM, to more accurately diagnose skin tumors and improve patient outcomes.
Innovation: UCLA researchers in the Departments of Electrical and Computer Engineering & Bioengineering have developed a deep learning-based framework to rapidly analyze in vivo RCM images. The method rapidly converts unstained skin images into virtually-stained hematoxylin and eosin-like (H&E) images with microscopic resolution. The network was trained to rapidly perform virtual histology of in vivo stacked RCM samples to generate an entirely biopsy-free method. Furthermore, the in vivo virtual staining method described can allow for an accelerated and accurate non-invasive diagnosis of skin tumors and improve clinical ability to detect malignant skin neoplasms.
Potential Applications: • Skin diagnostics • Diagnosis of malignant skin neoplasms • Cell analysis • Virtual histology of additional sample types Advantages: • Non-invasive • Rapid diagnostics • Accurate diagnostics • Biopsy-free • High resolution • Deep-learning based model • Integrates with current non-invasive imaging techniques
Development to Date: First successful demonstration of the biopsy-free virtual staining technique.