Potential of Machine Learning
HFpEF is heterogeneous entity affected by multiple clinical variables and structural alterations. Current methods resort to an oversimplified approach for assessment and risk stratification 4. This one size fits all approach has not proved applicable in clinical practice. With the evolution of technology and computer capabilities, artificial intelligence (AI) is opening new frontiers in cardiovascular imaging. Machine learning (ML), a subset of AI, is ushering a new era in cardiovascular imaging by expanding boundaries not limited by conventional statistics 49. Unlike traditional approaches, ML can decipher hidden patterns and extrapolate hidden patterns within vast data matrices 50,51. This technology can integrate diastolic indices and speckle tracking echocardiography to offer innovative insights into HFpEF52 (Figure 4).
Lancaster et al performed an unsupervised, hierarchical cluster analysis of 866 patients with diastolic dysfunction graded using contemporary ASE recommendations 53. Major adverse cardiovascular events, hospitalization, and mortality were compared between conventional and cluster-based categorizations. Survival analyses of patients assessed by clustering algorithms showed improved prediction of event-free survival by cluster analysis over diastolic grade classifications for all-cause mortality and cardiac mortality.
Omar et al performed a cluster analysis of LA and LV mechanical deformation parameters that resulted in a Doppler-independent phenotypic characterization of diastolic function and provided a noninvasive estimation of LV filling pressures 54. Speckle tracking features independently clustered patients into three groups with conventional parameters verifying increasing severity of dysfunction and LV filling pressure. Subsequent investigations from the same group lead to a refined ML model for assessing LV filling pressure using fourteen speckle tracking variables 55. This model correctly identified 80% of patients with pulmonary capillary wedge pressure ≥18 mm Hg,
ML has also been applied to resting and stress deformation imaging. Tabassian et al explored the role of ML in analyzing LV long axis mechanics during stress and rest in 100 patients including those with HFpEF and healthy, hypertensive, and breathless control subjects56. A ML algorithm was used to model spatiotemporal patterns of the speckle tracking traces and compare the ML algorithm predictions with the clinical diagnoses. The ML algorithm predicted symptoms with a high degree of accuracy and assigning subjects into four phenotypic groups. ML incorporating strain rate, compared with standard measurements, provided the greatest improvement in accuracy for predicting symptoms and 6-min walk distance. Sanchez-Martinez et al also utilized measurements of LV deformation at rest and exercise to examine differences between HFpEF and healthy patients from the MEDIA study (MEtabolic road to DIAstolic heart failure) 57. LV long-axis myocardial velocity patterns analyzed using an unsupervised ML algorithm identified a continuum from health to disease, including a transition zone associated with an uncertain diagnosis. As we move forward in the current era of cardiovascular imaging, ML algorithms will be increasingly integrated into clinical practice and cardiovascular research. These methods may help detect previously unrecognized phenotypes and tailor individualized therapies 4.