Leveraging Spectra to Improve the Resolution and Usability of Mid-Infrared Biomedical Microscopy: A Case Study in Bone Marrow Grading



Journal Title

Journal ISSN

Volume Title



Histological imaging is the current gold standard for disease diagnosis and prognosis. The current histological methods rely on techniques, such as chemical staining and pathological analysis, that are non-quantitative, subjective, and limited to specific biomarkers. Over the past decade, research has shown that spectroscopic imaging, when augmented with modern histology, may provide more precise measurements by overcoming variations in sample preparation and pathological expertise. The current mid-IR instrumentation though practical in a research environment, it is unsuitable for many clinical applications. Limitations of spectroscopic imaging include lengthy acquisition times, unmanageable data sizes, and diffraction limited spatial resolution. The goal of this dissertation is to study the application of mid-infrared (mid-IR) biomedical microscopy to improve usability in a clinical environment for histology. This is demonstrated by results for clinical analysis of bone marrow biopsies.

This dissertation has three parts. In the first part, I explore methods for optimal feature selection as it is critical for understanding chemical differences between tissue types as well as leveraging new discrete-frequency imaging systems. In the second part, I explore the potential for improving the spatial resolution of IR imaging systems by designing an image-fusion method based on the curvelet transform. Finally, I demonstrate the clinical viability of IR imaging for bone marrow grading. Through this research, I propose a roadmap and a set of requirements for efficient translation of infrared spectroscopic imaging into a clinical setting for bone marrow grading. The final results demonstrate that mid-infrared spectroscopic imaging has the potential to serve as a quantitative diagnostic tool for treatment tracking and disease progression with a case study of bone marrow fibrosis.



Spectroscopy, FTIR, Feature selection, Curveletes, Bone marrow, Machine learning


Portions of this document appear in: R. Mankar, M. Walsh, R. Bhargava, S. Prasad, and D. Mayerich, “Select-ing optimal features from fourier transform infrared spectroscopy for discrete-frequency imaging,”Analyst, vol. 143, pp. 1147–1156, 2018. And in: S. Ran, S. Berisha, R. Mankar, W.-C. Shih, and D. Mayerich, “Mitigatingfringing in discrete frequency infrared imaging using time-delayed integration,”Biomedical Optics Express, vol. 9, no. 2, pp. 832–843, 2018.