Back

AI/ML Models to Aid in the Diagnosis of COVID-19 Illness from Forced Cough Vocalizations: Results and Challenges of a Systematic Review of the Relevant Literature

Kelley, K.; Sakara, A. A.; Kelley, M.; Kelley, S. C.; McLenaghan, P.; Aldir, R.; Cox, M.; Donaldson, N.; Stogsdill, A.; Kotchou, S.; Sula, G.; Ramirez, M. A.

2021-11-14 public and global health
10.1101/2021.11.12.21266271 medRxiv
Show abstract

From a comprehensive and systematic search of the relevant literature on signal data signature (SDS)-based artificial intelligence/machine learning (AI/ML) systems designed to aid in the diagnosis of COVID-19 illness, we aimed to reproduce the reported systems and to derive a performance goal for comparison to our own medical device with the same intended use. These objectives were in line with a pathway to regulatory approval of such devices, as well as to acceptance of this unfamiliar technology by disaster/pandemic decision makers and clinicians. To our surprise, none of the peer-reviewed articles or pre-print server records contained details sufficient to meet the planned objectives. Information amassed from the full review of more than 60 publications, however, did underscore discrete impediments to bringing AI/ML diagnostic solutions to the bedside during a pandemic. These challenges then were explored by the authors via a gap analysis and specific remedies were proposed for bringing AI/ML technologies in closer alignment with the needs of a Total Product Life Cycle (TPLC) regulatory approach.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.