Gynecologic Cancer Tissue Analysis using Infrared Imaging
Introduction
Mid-infrared spectroscopy enables the identification of the biochemical constituents of tissue samples. The application of vibrational spectroscopy, particularly Mid-Infrared Spectroscopic Imaging (MIRSI), provides a significant advancement in the biochemical profiling of tissue sections, specifically for chemically heterogeneous samples such as cancerous tissues. Combining the molecular specificity of mid-infrared spectroscopy with the detailed spatial resolution of microscopy, MIRSI offers comprehensive chemical maps, thereby becoming instrumental in various fields, including medical diagnosis.
Methods
MIRSI, or Mid-Infrared Spectroscopic Imaging, has emerged as a pivotal tool in the field of biomedical imaging. Its most popular form, Fourier-Transform Infrared (FT-IR) Spectroscopic Imaging, has been instrumental in diagnosing a variety of cancers. With its ability to identify both the type and grade of cancer in a label-free, quantitative manner, MIRSI has replaced the need for multiple immunohistochemical stains. Furthermore, it opens the pathway for automated cancer diagnosis through the integration of chemometrics and machine learning tools. The introduction of new mid-IR sources, such as Quantum Cascade Lasers (QCLs), has further transformed the landscape, offering innovative imaging modalities with distinct advantages over FT-IR. Among these advancements, photothermal MIRSI techniques have enabled IR imaging at the nanoscale. However, the advent of these novel techniques also brings forth new challenges, necessitating the modeling and analysis of the underlying optical phenomena and instruments. We aim to tackle these complexities using models that enhance understanding of such instrumentation and the corresponding data. Further, we leverage a combination of these advanced instruments and machine learning techniques for precise cancer tissue segmentation. This integrated approach is applied for the diagnosis of a set of gynecologic cancers, including cervical, endometrial, and ovarian cancer, demonstrating the significant potential of MIRSI in cancer diagnosis and research.
Results
Gynecologic cancers represent one of the most lethal threats to women's health in the U.S., with early disease detection being pivotal for improving survival rates. To automate the diagnosis of ovarian, cervical, and endometrial cancers, we harness the capabilities of Mid-Infrared Spectroscopic Imaging (MIRSI) coupled with machine learning techniques. This approach, however, requires data of exceptional quality and resolution. Addressing this prerequisite, we utilize the super-resolution attributes of photothermal MIRSI for intricate ovarian tissue analysis and subsequent tissue subtype segmentation.
Our study showcases the robust and reliable imaging capabilities of novel instrumentation, exhibiting high signal-to-noise ratios (SNR) across 100 patient samples each for ovarian, cervical, and endometrial cancer. Furthermore, the employment of deep learning techniques enables us to achieve impressive tissue subtyping accuracy rates: 0.98 for ovarian, 0.94 for cervical, and 0.85 for endometrial cancer, illustrating the potential of this integrated approach for efficient cancer diagnosis.
Conclusion
The integration of molecular imaging and machine learning offers promising potential for enhancing the effectiveness of early diagnosis for ovarian, cervical, and endometrial cancers. We showcase a state-of-the-art instrumentation and data analysis pipeline devised to achieve this objective.