Abstract
Reporters obtained the following quote from the background information supplied by the inventors:“This invention relates generally to using artificial intellig ence (AI) to directly determine a patient specifictreatment or management of a disease that will benefit that patient, rather than manual selection basedon a diagnosis. Currently, autonomous AI systems use machine learning or other optimi zation techniquesto determine the diagnosis of a patient, and clinicians then u se this AI diagnosis, in addition to otherrelevant patient and population infor mation, to subjectively determine patient specific management orprescribe a pat ient specific treatment (also referred to herein as an “intervention”). Thus, Au tonomousAI diagnosis, while capable of determining a diagnosis without human ov ersight, still relies on a clinician tointerpret the diagnosis in terms of the patient’s entire case and then decide the intervention. However, thisintermedia te, subjective decision step is subject to high inter- and intra clinician varia bility, temporal andother drift. Moreover, the interaction between artificial i ntelligence and the clinician is variable with oftenunanticipated risks, and ha s been known to worsen rather than improve outcome. Moreover, obtainingthe high est quality reference standard (‘ground truth’) to train autonomous AI models fo r producinga diagnosis can be ethically problematic, and expensive. Where the r eference standard is dependent onclinician expertise, such as subjective readin g of images, the reference standard can be noisy. Instead,when clinical outcome , (which combines the effects of both the accuracy of the diagnostic process aswell as the precision of the treatment or management selection process) rather t han the interim diagnosiscan be used as reference standard for training the AI, this subjective, noisy step is eliminated, with thepotential to have higher pe rformance.”
NETL
NSTL