Histopathology Detection Using High-resolution Infrared Spectroscopic Imaging for Nodal Metastases in Oral Squamous Cell Carcinoma

Oral squamous cell carcinoma (OSCC) predominantly metastasises to lymph nodes and poses significant diagnostic challenges. The current gold standard for analysing OSSC biopsies is Hematoxylin and Eosin (H&E) staining, which provides essential tissue morphology information. Fourier Transform Infrared (FTIR) imaging can be used to complement (H&E) staining diagnosis. FTIR images provide information on chemical composition at diffraction-limited spatial resolution. This dissertation presents two novel approaches to overcome the spatial resolution limitations of FTIR imaging: a regression fusion model combining the high spatial resolution of (H&E) stains with the spectral information from FTIR and Optical Photothermal Infrared Micro-Spectroscopy (O-PTIR). The experiments utilised formalin-fixed, paraffin-embedded OSCC cervical lymph node metastases tissue microarrays (TMAs) with 1 mm diameter tissue cores. IR imaging was conducted using the Agilent Cary 620-FTIR imaging microscope, while O-PTIR micro-spectroscopy images were acquired in both reflection and transmission modes. The fusion models were employed to merge co-registered pairs of FTIR and H&E images, with the quality of fusion assessed using the structural similarity index measure (SSIM) and Spectral Angular Mapper (SAM). The results demonstrate minimal distortion and enhanced spatial resolution. Analysis of O-PTIR data in both reflection and transmission modes revealed that the reflection mode offered more detailed images with reasonable morphology and signal-to-noise ratio, while the transmission mode required higher laser power, posing potential sample damage risks. Ratio images from O-PTIR show contrast similar to H&E images. Pixel-wise models struggled to reproduce tissue discrimination, primarily due to information loss from neighbouring pixels. Substantial improvements in accuracy scores were achieved with a spatial-spectral model employing a hybrid convolutional neural network-random forest (CNN-RF) approach. In summary, this research demonstrates that image fusion techniques and O-PTIR can surpass the diffraction limits found in traditional FTIR techniques. Overall, this dissertation contributes to the advancement of IR molecular histopathology, particularly in the challenging context of imaging highly complex tissues such as OSCC metastases in lymph nodes.