RESUMEN
Oral potentially malignant disorders (OPMDs) are precursors to over 80% of oral cancers. Hematoxylin and eosin (H&E) staining, followed by pathologist interpretation of tissue and cellular morphology, is the current gold standard for diagnosis. However, this method is qualitative, can result in errors during the multi-step diagnostic process, and results may have significant inter-observer variability. Chemical imaging (CI) offers a promising alternative, wherein label-free imaging is used to record both the morphology and the composition of tissue and artificial intelligence (AI) is used to objectively assign histologic information. Here, we employ quantum cascade laser (QCL)-based discrete frequency infrared (DFIR) chemical imaging to record data from oral tissues. In this proof-of-concept study, we focused on achieving tissue segmentation into three classes (connective tissue, dysplastic epithelium, and normal epithelium) using a convolutional neural network (CNN) applied to three bands of label-free DFIR data with paired darkfield visible imaging. Using pathologist-annotated H&E images as the ground truth, we demonstrate results that are 94.5% accurate with the ground truth using combined information from IR and darkfield microscopy in a deep learning framework. This chemical-imaging-based workflow for OPMD classification has the potential to enhance the efficiency and accuracy of clinical oral precancer diagnosis.
RESUMEN
Chemical imaging, especially mid-infrared spectroscopic microscopy, enables label-free biomedical analyses while achieving expansive molecular sensitivity. However, its slow speed and poor image quality impede widespread adoption. We present a microscope that provides high-throughput recording, low noise, and high spatial resolution where the bottom-up design of its optical train facilitates dual-axis galvo laser scanning of a diffraction-limited focal point over large areas using custom, compound, infinity-corrected refractive objectives. We demonstrate whole-slide, speckle-free imaging in ~3 min per discrete wavelength at 10× magnification (2 µm/pixel) and high-resolution capability with its 20× counterpart (1 µm/pixel), both offering spatial quality at theoretical limits while maintaining high signal-to-noise ratios (>100:1). The data quality enables applications of modern machine learning and capabilities not previously feasible - 3D reconstructions using serial sections, comprehensive assessments of whole model organisms, and histological assessments of disease in time comparable to clinical workflows. Distinct from conventional approaches that focus on morphological investigations or immunostaining techniques, this development makes label-free imaging of minimally processed tissue practical.
