A low-cost machine learning-based cardiovascular/stroke risk assessment system: integration of conventional factors with image phenotypes

• DOI: 10.21037/cdt.2019.09.03
• Publication Date: 2019-10-22T07:47:47Z
• Authors: Ankush Jamthikar; Deep Gupta; Narendra N. Khanna; Luca Saba; Tadashi Araki; Klaudija Viskovic; Harman S. Suri; Ajay Gupta; Sophie Mavrogeni; Monika Turk; John R. Laird; Gyan Pareek; Martin Miner; Petros P. Sfikakis; Athanasios Protog...; George D. Kitas; Vijay Viswanathan; Andrew Nicolaides; Deepak L. Bhatt; Jasjit S. Suri
• Keywords: Risk Stratification

Background: Most cardiovascular (CV)/stroke risk calculators using the integration of carotid ultrasound image-based phenotypes (CUSIP) with conventional factors (CRF) have shown improved stratification compared either method. However such approaches not yet leveraged potential machine learning (ML). intelligent ML strategies use follow-ups for endpoints but are costly and time-intensive. We introduce an integrated system stenosis as endpoint training determine whether a can lead to superior performance system. Methods: The ML-based algorithm consists offline online extracts 47 features which comprised 13 CRF 34 CUSIP. Principal component analysis (PCA) was used select most significant features. These were then trained event-equivalent gold standard (consisting percentage stenosis) random forest (RF) classifier framework generate coefficients. transforms PCA-based test coefficients predict labels on subjects. above determines area under curve (AUC) 10-fold cross-validation paradigm. so-called "AtheroRisk-Integrated" against "AtheroRisk-Conventional", where only considered in feature set. Results: Left right common arteries 202 Japanese patients (Toho University, Japan) retrospectively examined obtain 395 scans. AtheroRisk-Integrated [AUC =0.80, P<0.0001, 95% confidence interval (CI): 0.77 0.84] showed improvement ~18% AtheroRisk-Conventional (AUC =0.68, CI: 0.64 0.72). Conclusions: model is powerful offers low cost high CV/stroke assessment.