Comparative modeling of mixed cardiopulmonary sounds in a low-resource paired dataset: Discrimination, calibration, and operating-point behavior

Runchen Cai · 2026

Read Comparative modeling of mixed cardiopulmonary sounds in a low-resource paired dataset: Discrimination, calibration, and operating-point behavior by Runchen Cai (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0352180.txt.

Description

Runchen Cai, writing solo, compared algorithms that must judge heart and lung health from a single mixed recording, the way a real stethoscope hears both at once. Most automated tools train on heart sounds or lung sounds separately, an easier problem than the bedside mixture. Using a small paired dataset that also holds the clean separate sources, Cai tested feature-based methods, multitask networks, and models taught by larger classic datasets.

A light model that could peek at the paired clean sources during training discriminated best overall, though scores swung widely between folds. Calibration and threshold behavior, not just accuracy, separated the contenders. On scarce, messy clinical data, the study suggests model choice should be argued at the operating point a clinic will actually use, and that extra paired recordings may beat bigger architectures.

Catalog record

CREATORRunchen Cai
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29183
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0352180

Browse the full library · About the Tower of Babel