I am a pediatric cardiologist and have cared for children with heart disease for the past three decades. In addition, I have an educational background in business and finance as well as healthcare administration and global health – I gained a Masters Degree in Public Health from UCLA and taught Global Health there after I completed the program.
There is a growing portfolio of technological tools that are available or soon to be so in the domain of artificial intelligence in medicine. The most advanced thus far appears to be deep learning in the form of convolutional neural networks (CNN) used in medical imaging in the fields of radiology and cardiology as well as ophthalmology, pathology, and dermatology. This Cambrian explosion of CNN tools have made steady progress in not only static imaging but also starting to make inroads into moving images such as ultrasound studies, endoscopic imaging, and even echocardiograms. AI has also been deployed in the workflow of medical image acquisition and decision making. Machine and even deep learning have also made progress in decision support with electronic medical records, but these have not been as productive as CNN and medical imaging due to the data being fragmented and inaccurate and medical decisions being complex in nature. There is also some promising machine and deep learning work in the area of drug design or repurposing in a myriad of diseases especially with protein structure determination based on genomic sequencing. Machine learning can also be used for patient identification for facilitating clinical trials and for real world data and experience.
Machine and deep learning in several forms are being adopted in healthcare. First, machine learning in the form of unsupervised learning holds great promise for discovery of new phenotypic expressions of disease subtypes and treatment responses. In addition, generative adversarial networks (GANs) are used for synthetic data in healthcare. The concept of transfer learning is also used increasingly more frequently in healthcare. Self-supervised and semi-supervised learning are also AI methodologies being explored in healthcare.
More recently, natural language processing (NLP) capabilities especially with transformer architectures such as Generative Pre-trained Transformer 3 (GPT-3) have started to be considered for their deployment in healthcare especially with their relative ease of use and expansive libraries. Cognitive computing, a more human-like variant of AI will continue to have presence in the panoply of AI in healthcare tools. In addition, the future prediction models will incorporate both structured data (vital signs, laboratory values, etc) and proportionally more unstructured data (clinician notes, radiology reports, etc.) so the models will have the best of both of these data and information worlds in the form of multimodal AI. This AI of incorporating many data streams will incorporate data from medical devices and sensors. Lastly, healthcare is starting to embrace an older AI technology of robotic process automation (RPA) for administrative tasks that can be automated by algorithms rather than completed by humans.