Validation of the artificial intelligence software to detect chronic fibrosing interstitial lung diseases in chest X-ray

Abstract

<b>Background:</b> Early use of anti-fibrotic drugs for idiopathic pulmonary fibrosis and progressive fibrosing interstitial lung diseases (ILD) is recommended to restrain deterioration of the respiratory function and prevent the acute exacerbation. However, it is challenging to find patients with ILD in their early stage with the chest X-ray examination. Recently, we developed an artificial intelligence (AI) software to help doctors detect chronic fibrosing ILD (CF-ILD). We currently validated the detection capability of this program using over 1200 chest X-ray images confirmed whether they have CF-ILD or not on the computed tomography (CT). <b>Method:</b> We identified consecutive chest X-ray images of patients who visited Sapporo Medical University Hospital from Jan 1 to Dec 31, 2019 and took chest X-ray and CT in the same day. We let the AI software interpret chest X-ray images and calculate the “likelihood score” for CF-ILD. CT images were interpreted by four specialists of pulmonary radiography and classified into normal, CF-ILD suspected/proven and having other abnormal findings. We investigated the detecting capability of AI software by AUC of ROC curve made from the sensitivity and false positive rate. <b>Result:</b> We identified 1280 patients and images, which included 352 with CF-ILD, 666 with other abnormal findings and 310 without abnormal findings. The AUC of ROC curve was 0.889 among CF-ILD-positive and CF-ILD-negative and 0.960 among CF-ILD-positive and no abnormal findings. <b>Conclusion:</b> Our AI software to detect CF-ILD in the chest X-ray showed a good detection performance.