Development of a Neural Network for Prediction of Macromolecular Composition in Hyperspectral Imaging Data Sets

The purpose of this acquisition is to purchase service of Development of a Neural Network for Prediction of Macromolecular Composition in Hyperspectral Imaging Data Sets. 

The MRI Section of the National Institute on Aging (NIA) Intramural Research Program (IRP) in studies of tissue response to aging and age-related pathology. Although the Section’s main current interest is applications within the central nervous system, the methodology developed is more wide-ranging and has potential utility across a broad range of problems in biomedical tissue characterization. The MRI Section has a strong interest in extending the Section’s methodology to other areas. In particular, the lab’s now propose to build on our published work on neural network analysis of biomedical signals to address the difficult problem of macromolecular quantification in near-infrared (NIR) hyperspectral imaging of cartilage and other tissues. NIR experiments are central to minimally invasive assessment of cartilage and other tissues at the molecular level and have been investigated by several groups in recent years. However, unique expertise is required to obtain hyperspectral images that span both the NIR and mid-infrared (MIR) region. The ability to obtain this data from individual pixels at high resolution in one scan enables the NIR hyperspectral data to serve as input to a neural network where the MIR data, arising from a gold-standard approach for obtaining macromolecular composition information, can serve as the outcome data.

The considerations above lead to a very natural formulation of a neural-network approach using hyperspectral images of cartilage that span both the NIR and MIR ranges. Using a wide-bandwidth spectrometer, NIR and MIR spectra will be jointly acquired at high pixel resolution (micron-level). The latter can then be quantified for component sizes of collagen and proteoglycan (PG). This will be performed in an imaging modality, where, in effect, each pixel will serve as a training sample of paired data of the form (NIR spectra; [collagen] and [PG]). After training, the NN can be applied to quantification of the NIR-only spectra obtained from animal studies, human subjects and patients; The NN can also serve as a template for similar approaches in other tissues. Application to newly acquired human data is outside of the scope of the present proposal but serves as the motivation and goal. 

Period of Performance: 9/15/2024-6/30/2025