Enthusiasm about the promises of machine learning applications in the biomedical field is evergrowing, pushed by the amazing list of SOTA established by data-driven computational approaches in recent years. Nevertheless, real-world applications are lagging behind, more so as we focus our attention closer on the patients and the final decisions to be taken during care, where problems like clinical validation (lack thereof), regulatory grey areas, and the challenges of bias moderation hit the hardest. Whilst advocates are especially vocal about data availability being the bottleneck for progress in such areas, some obstacles (e.g. challenges on the path to infer causality from data, the inevitability of the crystallisation of prior biases from data into the learnt models, lacklustre design research in communicating uncertainty and supporting evidence/ elements to a human user, …) are objectively not as trivial as the community initially framed them to be.
Interest in model-based computational approaches has been growing, and in some notable cases (e.g. Virtual Physiological Human) has been able to attract meaningful institutional support. The advantage of models is that they make the domain of existence/relevance and their fundamental assumptions explicit, allowing humans at once to learn from their failures and refinement and to confidently decide when/how to rely on them. Causality inference, and biases, are thus amenable to explicit conversations, and the approach has a more prominent place for humans in the loop compared to alternatives. On the downside though, model-based computational approaches require longer to approach new problems and can extend as far as our knowledge stretches.
Lead by an international team of subject experts, the editors call for papers inviting reflections on the strengths and weaknesses of both data-driven and model-based approaches, with special regard to the state of the art and adjacent future in cardiovascular medicine, and especially wish to receive submissions by authors invested in bridging the two worlds to ensure that the best of each is at the service of patients, caregivers, and healthcare systems.
Clinical trials, hypothesis & theory and original research (especially of the kind solidly rooted in theoretical work) will receive priority. Case reports are invited especially around near-miss/failures and dealing with ethical pitfalls, but will also be considered when proposing evidence directly relevant to future developments. Technology and code will be considered when particularly novel or of great potential impact in cardiovascular medicine. Policy and practice reviews will also be considered.
Enthusiasm about the promises of machine learning applications in the biomedical field is evergrowing, pushed by the amazing list of SOTA established by data-driven computational approaches in recent years. Nevertheless, real-world applications are lagging behind, more so as we focus our attention closer on the patients and the final decisions to be taken during care, where problems like clinical validation (lack thereof), regulatory grey areas, and the challenges of bias moderation hit the hardest. Whilst advocates are especially vocal about data availability being the bottleneck for progress in such areas, some obstacles (e.g. challenges on the path to infer causality from data, the inevitability of the crystallisation of prior biases from data into the learnt models, lacklustre design research in communicating uncertainty and supporting evidence/ elements to a human user, …) are objectively not as trivial as the community initially framed them to be.
Interest in model-based computational approaches has been growing, and in some notable cases (e.g. Virtual Physiological Human) has been able to attract meaningful institutional support. The advantage of models is that they make the domain of existence/relevance and their fundamental assumptions explicit, allowing humans at once to learn from their failures and refinement and to confidently decide when/how to rely on them. Causality inference, and biases, are thus amenable to explicit conversations, and the approach has a more prominent place for humans in the loop compared to alternatives. On the downside though, model-based computational approaches require longer to approach new problems and can extend as far as our knowledge stretches.
Lead by an international team of subject experts, the editors call for papers inviting reflections on the strengths and weaknesses of both data-driven and model-based approaches, with special regard to the state of the art and adjacent future in cardiovascular medicine, and especially wish to receive submissions by authors invested in bridging the two worlds to ensure that the best of each is at the service of patients, caregivers, and healthcare systems.
Clinical trials, hypothesis & theory and original research (especially of the kind solidly rooted in theoretical work) will receive priority. Case reports are invited especially around near-miss/failures and dealing with ethical pitfalls, but will also be considered when proposing evidence directly relevant to future developments. Technology and code will be considered when particularly novel or of great potential impact in cardiovascular medicine. Policy and practice reviews will also be considered.
