D

Every day, it seems that some new algorithm(算法)enables computers to disease a disease with unprecedented accuracy, renewing predictions that computers will soon replace doctors. What if computers could replace patients as well? If virtual humans could have replaced real people in some stages of a coronavirus vaccine trial, it could have sped development of a preventive tool and slowed down the pandemic. Similarly, potential vaccines that weren't likely to work could have been identified early, reducing trial costs and avoiding testing poor vaccine candidates on living volunteers. These are some of the benefits of "in silico medicine", or the testing of drugs and treatments on virtual organs or body systems to predict how a real person will respond to the therapies.

The modeling begins by feeding anatomical data drawn from noninvasive(非侵入式)high-resolution imaging of an individual's actual organ into a complex mathematical model of the mechanisms that govern that organ's function. Algorithms running on powerful computers resolve the resulting equations and unknowns, generating a virtual organ that looks and behaves like the real thing.

In silico clinical trials are already underway to an extent. HeartFlow Analysis, for instance, enables clinicians to identify CAD（冠心病） based on CT images of a patient's heart. The Heart Flow system uses these images to construct a fluid dynamic model of the blood running through the coronary blood vessels, thereby identifying abnormal conditions and their severity. Without this technology, doctors would need to perform an invasive operation to decide whether and how to intervene. Experimenting on digital models of individual patients can also help personalize therapy for any number of conditions and s already used in diabetes care.

The philosophy behind in silico medicine is not new. The ability to create and simulate the performance of an object under hundreds of operating conditions has been a cornerstone of engineering for decades, such as for designing electronic circuits, airplanes and buildings. Various obstacles remain to its widespread implementation in medical research and development.

The predictive power and reliability of this technology must be confirmed, and that will require several advances. Those include the generation of high quality medical databases from a large, ethnically diverse patient base that has both women and men; improvement of mathematical models to account for the many interacting processes in the body; and further modification of AI methods that were developed mainly for computer-based speech and image recognition and need to be extended to provide biological insights.

In recent years American and European regulators have approved some commercial uses of computer-based diagnostics, but meeting regulatory demands requires considerable time and money. Creating demand for these computer-based diagnostic tools is challenging as well. In silico medicine must be able to deliver cost-effective value for patients, clinicians and health care organizations to accelerate their adoption of the technology.