AI can identify COVID-19 in lungs like physicians can, study shows

” We demonstrated that a deep learning-based AI approach can work as a objective and standardized tool to assist health care systems along with patients,” Ulas Bagci, PhD, one of the studys researchers, stated in a Sept. 30 news release. “It can be utilized as a complementary test tool in extremely particular limited populations, and it can be used rapidly and at big scale in the regrettable event of a recurrent break out.”.

The research study team trained an expert system algorithm to find COVID-19 in lung scans of 1,280 clients from Japan, China and Italy. They checked it on 1,337 patients with lung illness spanning from COVID-19 to pneumonia and cancer.

More posts on artificial intelligence: 4 members of Congress need research on racism within clinical algorithmsDefense Department, Philips broaden AI research for early detection of transmittable diseasesBulk of clinical AI systems data originates from just 3 states, research study discovers.

© Copyright ASC COMMUNICATIONS 2020. Intrigued in LINKING to or REPRINTING this content? View our policies by clicking here.

Register today for Beckers HIT+RCM Virtual Event Oct. 6-9 for the best insights and big concepts in health IT!

The research study was performed to check out alternatives to reverse transcription-polymerase domino effect tests, which are often used to diagnose COVID-19. These tests often go through hold-ups during processing and have a high risk of producing incorrect negatives.

© Copyright ASC COMMUNICATIONS 2020. Intrigued in LINKING to or REPRINTING this material? View our policies by click on this link.

An algorithm can determine COVID-19 cases and separate them from influenza cases with the same precision as a doctors diagnosis, according to a study just recently published in Nature Communications.

The algorithm had the ability to precisely diagnose 84 percent of positive COVID-19 cases and 93 percent of negative cases.

Katie Adams –
Wednesday, September 30th, 2020
Print|Email.