New perspectives in digital medicine: How artificial intelligence diagnoses of bacterial vaginosis differ

(Elana Gotkine – HealthDay News) – The performance of a machine learning (ML) model in diagnosing bacterial vaginosis (BV) differs between ethnic groups, according to a study published online Nov. 17 in npj. Digital Medicine.

Cameron Celeste of the University of Florida in Gainesville and colleagues examined the ability of four ML algorithms to diagnose bacterial vaginosis. The fairness in predicting asymptomatic BV was investigated using 16S rRNA sequencing data from Asian, black, Latina, and white women.

Researchers have observed differences in the performance of general-purpose machine learning models based on race. Models performed less effectively for Latina and Asian women when assessing the rate of false positive (FP), that is, healthy women who appeared sick, or false negatives (FN), corresponding to women with bacterial vaginosis who appeared healthy according to the AI.

Overall, the models had the highest and lowest performance for white and Asian women, respectively.

“Here, we show that several supervised learning models perform differently for ethnic groups by evaluating commonly used measures, such as balanced accuracy and mean accuracy, as well as clinically relevant measures, such as FP and FN, in a group of asymptomatic women. BV “, write the authors. “The results provide evidence of a discrepancy in typical performance between races.”