AI Case Study
Maria Fareri Children's Hospital deploys a diagnosis platform using NLP and algorithmic search to assist physicians with difficult to diagnose symptoms
A diagnostic assistant platform called Isabel uses natural language processing and machine learning to offer possible diagnoses for physicians. Symptoms are entered into Isabel as text and the system's algorithms search through its medical database to offer likeliness of different diagnostic predictions, with claimed 95% accuracy.
Healthcare Providers And Services
As reported in the Hudson Valley Business Journal, "Maria Fareri
Children’s Hospital at Westchester Medical Center is the first children’s hospital in New York State to provide pediatric specialists with an additional resource in diagnosing complex conditions. The Isabel system is a massive web-based medical database. As a tool, it allows physicians to interact in real-time
providing immediate reference materials based on a patient’s symptoms and clinical issues." From the Isabel website: "Initially designed for childhood illnesses, the DDx [differential diagnosis] generator was expanded in 2006 to handle both adult and pediatric conditions and then, in 2012, a modified version was released as a free to use Symptom Checker for patients. It is supposed to offer a second pair of eyes."
Often times rarer diseases are misdiagnosed as more common ones and can have severe consequences. The idea behind Isabel was to offer all diagnostic possibilities and started out as a database, eventually evolving to using AI to streamline predictions and natural language processing.
No specific results presented for the Maria Fareri Children's Hospital, however studies have shown the Isabel system has 95% accuracy in predictions.
Risk reduction - Predictive diagnosis
According to the Isabel website, the DDx works thanks to "innovative, statistical natural language processing software which understands the meaning and context of unstructured free text. This is applied to our medical database of disease presentations which comprises thousands of carefully selected, evidence-based documents describing the multiple ways that over 10,000 conditions can present. The engine has effectively been trained over almost two decades on virtually every known possible presentation of the majority of all known diseases. The initial results from the Isabel DPREP are then passed through an additional set of algorithms tuned over many years to ensure that only those results relevant to the patient’s age, gender and geographical region are displayed."
Input into Isabel is text, and the search itself is run on "thousands of carefully selected, evidence-based documents describing the multiple ways that over 10,000 conditions can present" according to the website.