Classification of lung sounds for the detection of interstitial lung disease secondary to rheumatoid arthritis

Rheumatoid arthritis is an autoimmune disease impacting around 1% of population. One of the most severe comorbidity is interstitial lung disease. Currently, the treatment is effective only in the very early stages of the disease. Symptoms appear late in the clinical history and are not useful for diagnosis. A routine use of high resolution computer tomography for screening programs is not advisable for both exposition to ionizing radiation of patients and high costs to be sustained by the national health system. Lung auscultation can reveal the so called “velcro crackle” associated to different radiological patterns. In this work we propose a new pipeline for pre-processing and classification of lung sounds suitable to the detection of interstitial lung disease in patient affected by rheumatoid arthritis. The data set has been collected in a clinical study at the university hospital of Modena (Italy). Ground truth is represented by the high resolution computer tomography report. Accuracy and F1-score of our solution are 83.2% and 77,9% in the classification of lung sounds, respectively. Combining the predictions of the classifier for distinct auscultations of the same patient, the accuracy and F1-score get as high as 87.8% and 87,1%, respectively. Considering that physical lung auscultation is safe for the patient and cheap for the national health system, the proposed solution can pave the way for a screening campaign aimed at the early detection of interstitial lung disease secondary to rheumatoid arthritis.