Acoustic Epidemiology of Pulmonary Tuberculosis (TB) & Covid19 Leveraging explainable AI/ML

Involuntary cough is a prominent symptom for many a Lung Ailments ranging from Infectious to non-Infectious diseases. Early research around human cough established that the spectral signatures do not vary between Involuntary and Voluntary coughs. The study aimed at evaluating voluntary human cough sounds recorded under a stringent clinical protocol. India’s ambitious goal to eliminate and eradicate TB by 2025 shall be facilitated by Machine Learning tools that address subjectivity in that the healthcare worker can now take the solution as a screening modality to the last mile as a part of outreach programs without having to rely on infrastructure & connectivity. In this paper we present the findings of Clinical Trials for Pulmonary TB registered at CTRI/2019/02/017672 conducted independently and included Covid19 during the pandemic as a part of Bi-Directional screening modality. The reference standards used were CBNAAT (Cartridge based nucleic acid amplification test) & CXR (Chest X-Ray) for TB while for Covid19; RT-PCR was used as the reference standard. As a non-invasive and contactless screening modality, a sophisticated third-party Microphone Array was used to record the cough under a stringent infection control protocol. Sensitivity achieved across the sites for TB ranged between 80% - 83% and Specificity was to the tune of 92% while using CBNAAT as a reference standard. CXR when used as a reference standard for TB achieved a sensitivity and specificity of 59% and 60% respectively. Covid19 achieved a sensitivity & specificity of 92% and 96% while using RT-PCR as the reference standard. The study was primarily focused on the Frequency domain that paved way for feature extraction and explainable Machine Learning Models operating upon lossless WAV files hypothesizing acoustic theory and demographic inputs. The solution titled “TimBre” can now be added to the healthcare workers arsenal in situations where a RT-PCR or CXR is not available and seamlessly conduct bidirectional screening with a single recording of cough and also offer insights into Non-Communicable diseases as a part of differential diagnosis