Artificial intelligence models demonstrated high diagnostic accuracy for heart failure subtypes (e.g., sensitivity 0.99 and NPV 0.98 for HFpEF detection by DL-CNN ECG model) and achieved prognostic prediction performances with C-statistics up to 0.91 in specific cohorts, outperforming traditional methods but requiring further validation.
Abstract Background Heart failure (HF) affects millions of individuals worldwide and shows an increasing trend, constituting a serious public health issue. Considerable attention has been paid to the screening, diagnosis, risk prediction, treatment, and prognosis of HF. Although many guidelines for the management of HF have been proposed in recent years, the efficacy of evidence-based treatments seems to vary among patients. Therefore, the era of “one-size-fits-all” approaches is drawing to a close, and the concepts of precision medicine and individualized medicine are gradually taking root. Artificial intelligence (AI) is an emerging discipline in the rapidly growing field of computer science. It has now become deeply involved in all aspects of cardiovascular disease research, with particular relevance to HF, though its translation into clinical practice is yet to be fully realized. Although the use of AI in cardiovascular disease (CVD) and HF patient care, as well as cardiac resynchronization therapy (CRT), has been extensively discussed, a discussion from the standpoint of all aspects of HF clinical process is lacking. Main body This review provides a comprehensive overview of the use of AI in HF in specific scenarios, including patient diagnosis, subtyping, prognostic assessment, pre- and post-treatment evaluation, and telecare. It also presents the prospects and challenges for the development of AI in the field of HF, with the expectation that a mature AI diagnosis and treatment system adapted to clinical practice will be developed in the future through in-depth research and validation. Conclusions This review summarizes the application of AI in various links of HF management from diagnosis to telecare, and analyzes its current application limitations, existing challenges and future research directions, aiming to provide a reference for the subsequent clinical transformation and research optimization of AI in the HF field.
Li et al. (Sun,) conducted a review in Patients with heart failure including subtypes HFpEF, HFmrEF, and HFrEF in various clinical settings. Artificial intelligence-based diagnostic and prognostic models vs. Traditional clinical scoring systems or physician expert judgment was evaluated on Diagnostic or prognostic accuracy for heart failure diagnosis and outcomes prediction. Artificial intelligence models demonstrated high diagnostic accuracy for heart failure subtypes (e.g., sensitivity 0.99 and NPV 0.98 for HFpEF detection by DL-CNN ECG model) and achieved prognostic prediction performances with C-statistics up to 0.91 in specific cohorts, outperforming traditional methods but requiring further validation.