Label-Free Cancer Classification Using Hyperspectral Imaging Microscopy and Machine Learning

WHO reported breast cancer as the most prevalent cancer in the world, with 2.26 million cases in 2020 [1]. Current methods for cancer diagnosis involve the pathologist's analysis of a stained biopsy tissue slide viewed under a brightfield microscope, which presents an inherent time delay due to staining, and diagnosis is limited to the pathologist's judgments. To overcome these challenges, we used darkfield and phase contrast microscopy technology because they are suitable for imaging thin and transparent samples. To perform cancer diagnosis automatically, we used hyperspectral imaging (HSI) technology combined with machine learning because HSI captures images in hundreds of spectral channels, providing spectral and spatial information about the sample. We developed machine learning models and demonstrated a high classification accuracy using label-free unstained slides. The results are comparable to those obtained using the hematoxylin and eosin (H&E) stained slides. Hence our approach may improve the current histopathology procedure in cancer diagnosis.