A key evolution and need increasingly recognized in the realm of healthcare AI is the customization of algorithms dedicated to the unique problems posed by medicine. By contrast, most current work in this field can be distilled as an applied science, whereby existing algorithms published by Google and Facebook are used (somewhat indiscriminately) on healthcare problems. At Avicenna, we feel that a central tenet to our vision is the need to develop and drive research towards AI solutions customized for medical imaging. To do so, we emphasize key collaborations with physicians and other clinicians to ensure that trained models “think” and “reason” in ways that mimic their human counterparts.
In addition to algorithm development itself, another key ingredient that is needed to be successful is understanding how to scale the research and development process. After creating a paradigm or approach that solves one specific disease process, the same ideas can often times be applied to numerous additional similar pathologic entities with minor modifications. Furthermore, the clinical implementation of algorithms is another key, oftentimes forgotten component of design efficacious AI tools. This includes understanding how a powerful piece of software can be used effectively in an existing paradigm for patient care, and in what ways traditional healthcare workflow may need to be adapted to facilitate new technology. Ultimately, as before, engaging physicians and end-users is key to ensure that all these components of successful algorithm development are addressed in a meaningful way.
In the near future, I anticipate that the learning curve for developing new algorithms will become increasingly reduced; this in part is due to the release of popular and powerful open-source libraries, the accessibility of learning materials (including my own publicly available curriculum) and the sharing of large multi-institutional datasets. In this context, an anticipated role that leading commercial entities may need to embrace is to support and promote the growing population of algorithm developers for example within university hospitals, academic research departments or even the lay hobbyist. The role of companies like ours could shift from primary algorithm creation to one of facilitating discovery by providing uniform platforms to train, validate and deploy AI models, or perhaps to offer cloud-based services implementing complex AI software stacks for hospitals that do not have local technical expertise.
In any case, at Avicenna we recognize that the landscape of AI in healthcare is constantly evolving. As a result we will work tirelessly to adapt our vision to maximize the impact we may have and to continue pushing the field to new limits.