Machine learning integrated laser speckle image analysis for the simultaneous extraction of flow and scatterer concentration from capillary phantoms

Abstract

Laser speckle imaging is one of the powerful non-invasive imaging techniques to monitor and assess microcirculation parameters. Qualitative analysis of perfusion parameters has been carried out in the recent past. But the quantitative estimation of tissue perfusion parameters like flow velocity and scatterer concentration simultaneously from laser speckle images remains challenging. The introduction of machine learning methods into laser speckle image analysis can help meet these challenges to a great extent. This paper presents an approach for the simultaneous extraction of perfusion parameters, using multi-target regression techniques applied to the extracted features from acquired laser speckle images after Eigendecomposition filtering. The multi-target regression trees are identified as an effective tool for the simultaneous extraction of flow velocity and scatterer concentration with adequate mean absolute percentage error. Besides the achieved speed and computational efficiency, our work demonstrates the viability of this approach in quantifying perfusion parameters simultaneously. Due to its simple, non-invasive, and cost-effective nature, the proposed technique could be used in the realtime assessment of tissue health.