By Andrea Park, Becker’s Hospital Review | June 24, 2019
Here are seven recent studies and launches illustrating novel applications — some theoretical and some already in use — for machine learning artificial intelligence in clinical settings:
1. A machine and deep learning algorithm from scientists at the IBM Research laboratory in Haifa, Israel, analyzes health records and mammograms to predict the development of breast cancer up to one year before its onset with nearly 90 percent accuracy.
2. Researchers from the University of Pennsylvania’s Perelman School of Medicine and Stony Brook University developed AI that analyzes the language used in Facebook posts to detect conditions such as diabetes, anxiety, depression and psychosis, giving healthcare providers insights they may not be able to glean from a typical clinical consultation.
3. A new machine learning method trained by scientists at the University of Virginia to examine duodenal biopsy imaging data expedites the process of imaging, diagnosing and treating two types of pediatric gut disease.
4. Emory University scientists created an algorithm that detects patterns in an individual’s speech that can indicate the onset of the prodromal phase of psychosis with more than 90 percent accuracy.
5. A group of algorithms trained using a set of more than 10,000 dermatoscopic images initially developed for 2018’s International Skin Imaging Collaboration challenge outperforms dermatologists, dermatology residents and general practitioners in diagnosing pigmented skin lesions.
6. Engineers from Rensselaer Polytechnic Institute and radiologists from Massachusetts General Hospital and Harvard Medical School developed a machine learning model that produces low-dose CT images more quickly and with greater accuracy than past attempts to decrease the amount of radiation used in CT imaging.
7. University of Maryland Medical Systems deployed the Baltimore score, a proprietary model that predicts how likely discharged patients are to be readmitted based on nearly 400 variables, including demographics, lab test results, body mass index, church affiliation, marital status, employment and medication usage.