Small data in model calibration for optical tissue phantom validation

Abstract

Machine learning algorithms traditionally rely on large datasets for high accuracy. However, advances in the field are now enabling the exploration of solutions in niche engineering areas with smaller datasets. This article reviews the challenges and solutions in working with small datasets, particularly in optoelectronics and biomedical engineering. In optoelectronics, small datasets are key for designing and validating photonic systems, as experiments with living tissues can be costly and complex. The article discusses optimizing photonic response simulations and system calibration using machine learning models that are effective with smaller datasets. In biomedical engineering, the focus is on 3D-printed tissue phantoms, which mimic living tissue properties for non-invasive validation of photonic devices in diagnostics. The study explores how small data techniques like transfer learning, bootstrapping, regularization, and K-fold cross-validation can improve interpretations from small datasets, enhance predictive capabilities, and address data scarcity issues.