We are witnessing a new era in which artificial intelligence is being applied to areas of research, imaging, diagnostics, and treatment. In cardiovascular medicine, it is being used in various ways from genomics to cardiac imaging analysis, yielding technology and tools that could potentially change diagnostic testing to improve patient care. Thomas Stuckey explores a new approach that is using cardiac phase space tomography analysis and advanced machine learning to detect significant coronary artery disease.
The current methods of detecting coronary artery disease are cumbersome, often taking weeks or months for a diagnosis, and can expose patients to radiation and stress; and in some cases, unnecessary and invasive heart catheterisations. Manesh Patel (Duke Heart Center, Durham, USA) and colleagues report in The New England Journal of Medicine that up to two thirds of patients who are undergo invasive cardiac catheterisations are subsequently found not to have significant obstructive disease.1 As healthcare reimbursement, in the USA, moves into a new value-based healthcare model with capped coverage for patients, there is a timely need to find a new coronary artery disease testing pathway for both patients and physicians.
The CADLAD study
At the LeBauer-Brodie Center for Cardiovascular Research and the Cone Health Heart and Vascular Center, my colleagues and I have been conducting a clinical trial—CADLAD (Coronary artery disease learning and algorithm development)—to evaluate a novel imaging technology. Analytics 4 Life, the company behind the trial, is using artificial intelligence to develop an imaging technology to assess for the presence of coronary artery disease. This technology is designed to use only intrinsic phase signals scanned from the body; therefore, avoiding exposing patients to radiation, heart rate acceleration, or injections of contrast agents.
The CADLAD study is comparing the accuracy of Analytics 4 Life’s cardiac phase space tomography analysis (cPSTA) system to detect the presence of significant coronary artery disease to that of cardiac catheterisation results. The system uses a hand-held digital instrument that non-invasively scans and transmits a patient’s phase signal data to a secure, cloud-based repository, where software then analyses the data to identify areas of ischaemia from coronary artery disease that can be interpreted by physicians.
In stage I of the CADLAD trial, 606 phase signals were obtained from patients at rest, just prior to angiography. Angiographic results were paired with signal data, from which features were extracted, and a training set of 512 signals were used for machine learning, with 94 reserved for blind testing. In the cohort of 94 patients tested, the area under the curve (AUC) was 0.80. The study revealed the machine-learned predictor had a sensitivity of 92% (95% CI: 74%-100%) and specificity of 62% (95% CI: 51%-74%) on blind testing. The negative predictive value was 96% (95% CI: 85%-100%).
Gender-based assessment of coronary artery disease by cPSTA revealed positive initial data to suggest that resting cPSTA imaging performs well overall, and that the results in women are equivalent or better as compared with men. These findings are valuable for the historically under-served and difficult-to-diagnose female subpopulation whose nuclear scans may have breast imaging artifacts that can result in false positives.
Assessing coronary artery disease by cPSTA in elderly and obese patients, other difficult-to-diagnose subpopulations, performed well with negative predictive values of 83% (95% CI: 50%-100%) and 94% (95% CI: 79%-100%), respectively. The current coronary artery disease detection pathways for obese patients are at high-risk for abnormal test results, due to exercise limitations, diaphragmatic attenuation artifacts, and image acquisition limitations. These early results show a promising potential for alternative technology to establish new coronary artery disease detection pathways on specific populations where current conventional CAD detection pathways tend to be less accurate. More promising, no safety issues were observed (no AEs in 606 procedures).
Conclusion
Current CADLAD stage I results were in a population of high-risk coronary artery disease patients already showing signs of coronary artery disease, such as chest pain. The next stage will be to test for coronary artery disease in a low-risk population to prove continued success in the negative predictive value of the test. This innovative approach to cardiac diagnostics is a great example of how to leverage artificial intelligence in healthcare to improve patient care. There is no question that it will take time to adapt to evolving and disruptive technologies that artificial intelligence offers, but it is important that we keep an open mind and embrace change in order to improve patient care.
Thomas Stuckey is medical director at LeBauer-Brodie Center for Cardiovascular Research and Education at Cone Health, Greensboro, USA.
Reference
- Patel et al. N Engl J Med 2010; 362: 886–95.
