Leveraging high-resolution spatial features in mid-infrared spectroscopic imaging to classify tissue subtypes in ovarian cancer

Mid-infrared spectroscopic imaging (MIRSI) is an emerging class of label-free techniques being leveraged for digital histopathology. Optical photothermal infrared (O-PTIR) is based on vibrational absorbance imaging using a pump-probe architecture capable of a 10x enhancement in spatial resolution relative to FTIR imaging. This allows truly sub-cellular spectroscopic investigation of tissue at biochemically important fingerprint wavelengths. Modern histopathologic identification of ovarian cancer involves tissue staining followed by morphological pattern recognition. This process is time-consuming, subjective, and requires extensive expertise. In this paper, we present the first label-free automated histological classification of ovarian tissue sub-types using MIRSI. We demonstrate that enhanced resolution of sub-cellular features, combined with spectroscopic information, enables reliable classification (0.98 AUC) of ovarian cell sub-types. Moreover, we present statistically robust validation from 74 patient samples with over 60 million data points. This demonstrates that sub-cellular resolution from five wavenumbers is sufficient to outperform state-of-the-art diffraction-limited techniques from up to 374 different wavenumbers. O-PTIR also performs measurements in back-reflection geometry, opening the door to future in vivo studies on glass slides.