Tuberculosis Screening from Cough Audio: Baseline Models, Clinical Variables, and Uncertainty Quantification

Researchers have established a new standardized framework for detecting Tuberculosis (TB) using machine learning on cough audio and basic clinical data. The work addresses a major problem in the field: inconsistent methods make it impossible to compare results. They built a reproducible pipeline that fuses audio and metadata, performs uncertainty quantification, and reports consistent clinical metrics. The full protocol is released to serve as a common baseline, aiming to accelerate reliable progress in AI-powered disease screening.