Maxeler Technologies (London, UK), a leader in optimization of computational data movement, is collaborating with Professor Ophir Frieder at Georgetown University (Washington DC, USA) to commercialize Georgetown AI research aimed at reducing patient adverse effects related to errors in drug prescriptions, on Maxeler’s M-Space Software Delivery Platform. The novel AI prescription prediction technology licensed from Georgetown University could be utilized in hospitals, pharmacies, or in medical emergencies in the field. Using the licensed technology, Maxeler will work with Georgetown researchers to productize scalable, patient-specific prescription selectors, thereby reducing the risk of prescription errors and improving patient care.
[ClickPress, Thu Jan 23 2020] Maxeler Technologies recently licensed novel, “explainable” AI-driven prescription prediction technologies from Georgetown University. Using the patent-pending, technology developed at Georgetown’s Information Retrieval Laboratory under Professor Ophir Frieder, Maxeler intends to work with the researchers at Georgetown to productize scalable, patient-specific, prescription selectors, reducing drug resistance, thereby improving patient care. Frieder, the Robert L. McDevitt, K.S.G., K.C.H.S. and Catherine H. McDevitt L.C.H.S. Professor in Computer Science and Information Processing, is the lead inventor of the technology.
The AI-driven prescription predictive models suggest effective medications that minimize adverse effects. Many artificial intelligence based efforts focus on the aforementioned problem but do so using “black box” approaches which solve the problem, but fail to provide interpretability. Since prescribing physicians are hesitant to rely on suggestions whose derivations are not understood, the lack of interpretability renders the results of the black box AI unuseable. Frieder and his collaborators solved that problem by making the recommendations explainable.
High dimensional information and temporal event relationships complicate the development of predictive models. Traditional approaches transform and “flatten” electronic health records into vector representations that ignore medical event temporal relationships, reducing prediction accuracy. The explainable approach automatically restructures each patient’s electronic medical record into a graph and utilizes a graph-kernel approach to formulate prescription predictions. As graphs are easily understood both by doctors and patients, the developed system will provide comprehensible explanations of why a particular medication is prescribed. The products are expected to address domestic and international markets.
Maxeler intends to propose using medical records from a national health service to start adapting the research models to real-world data. “Our goal is to develop a model to process electronic medical records relying on proven, scalable, data-mining techniques, yielding a nearly real-time, personalized, clinically explainable drug prediction approach,” says Oskar Mencer, CEO of Maxeler Technologies. Antibiotic prescriptions will be the initial focus. This is because real-time personalized prescriptions are needed as resistance to antibiotics is personalized and develops over time. As Frieder explains, “This partnership with Maxeler provides us with the opportunity to move our research from the abstract to clinical practice, hopefully globally improving patient care.”
Read more here: http://maxeler.com/news.html
About Maxeler Technologies
Maxeler Technologies is one of the world’s leading companies with tools and services for software transformation, optimization of data movement and data representation, in High Performance Computing on standard CPUs as well as computing hardware for specific mission critical domains. Maxeler solutions including Cybersecurity, AI, Risk and Imaging have been used in production in Finance, Oil-and-Gas, Government and Academic Research.