Improving the Quality of Point of Care Diagnostics with Real-Time Machine Learning in Low Literacy LMIC Settings

The scalability of medical technology in low resource settings requires a higher level of usability and clear decision support compared to conventional devices, since users often have very limited training. In particular, it is important to provide users with real time feedback on data quality during the patient information acquisition in a manner that enables the user to take immediate corrective action. In this work, we present an example of such a system, which provides real time feedback on the source of noise and interference on a low cost Doppler device connected to a smart-phone used by traditional birth attendants (TBAs) in rural Guatemala. A total of 195 fetal recordings made from 146 singleton pregnancies in the second and third trimester were recorded over 8 months by 19 TBAs. The resulting 33.7 hours of data were segmented into 0.75 s epochs and independently labeled by three trained researchers into one of five noise or quality categories that dominated the data. A two-step classifier, composed of a logistic regression and a multiclass support vector machine, was then trained to classify the data on epochs from 0.75 s to 3.75 s. After feature selection the highest micro-averaged test F1 score was 96.8% and macro-average F1 test score was 94.5% for 3.75 s segments using 23 features. A reduced real time model with 17 features produced comparable micro-and macro-averaged test F1 scores of 96.0% and 94.5% respectively. The code is portable back to a low-end smartphone to run on such a device in real time (under 400 ms) in order to provide an audiovisual cue for the TBAs via the smartphone. Future work will evaluate the classifier presented here as part of a decision support system for data quality improvement in an ongoing randomized control trial in Guatemala.