AUTHOR=Kotronias Rafail A. , Fielding Kirsty , Greenhalgh Charlotte , Lee Regent , Alkhalil Mohammad , Marin Federico , Emfietzoglou Maria , Banning Adrian P. , Vallance Claire , Channon Keith M. , De Maria Giovanni Luigi TITLE=Machine learning assisted reflectance spectral characterisation of coronary thrombi correlates with microvascular injury in patients with ST-segment elevation acute coronary syndrome JOURNAL=Frontiers in Cardiovascular Medicine VOLUME=9 YEAR=2022 URL=https://www.frontiersin.org/journals/cardiovascular-medicine/articles/10.3389/fcvm.2022.930015 DOI=10.3389/fcvm.2022.930015 ISSN=2297-055X ABSTRACT=Aims

We set out to further develop reflectance spectroscopy for the characterisation and quantification of coronary thrombi. Additionally, we explore the potential of our approach for use as a risk stratification tool by exploring the relation of reflectance spectra to indices of coronary microvascular injury.

Methods and results

We performed hyperspectral imaging of coronary thrombi aspirated from 306 patients presenting with ST-segment elevation acute coronary syndrome (STEACS). Spatially resolved reflected light spectra were analysed using unsupervised machine learning approaches. Invasive [index of coronary microvascular resistance (IMR)] and non-invasive [microvascular obstruction (MVO) at cardiac magnetic resonance imaging] indices of coronary microvascular injury were measured in a sub-cohort of 36 patients. The derived spectral signatures of coronary thrombi were correlated with both invasive and non-invasive indices of coronary microvascular injury. Successful machine-learning-based classification of the various thrombus image components, including differentiation between blood and thrombus, was achieved when classifying the pixel spectra into 11 groups. Fitting of the spectra to basis spectra recorded for separated blood components confirmed excellent correlation with visually inspected thrombi. In the 36 patients who underwent successful thrombectomy, spectral signatures were found to correlate well with the index of microcirculatory resistance and microvascular obstruction; R2: 0.80, p < 0.0001, n = 21 and R2: 0.64, p = 0.02, n = 17, respectively.

Conclusion

Machine learning assisted reflectance spectral analysis can provide a measure of thrombus composition and evaluate coronary microvascular injury in patients with STEACS. Future work will further validate its deployment as a point-of-care diagnostic and risk stratification tool for STEACS care.