Development and external validation of a deep learning-based computed tomography classification system for COVID-19
Rationale: Currently available machine learning models for diagnosing COVID-19 based on computed tomography (CT) images are limited due to concerns regarding methodological flaws or underlying biases in the evaluation process. Objectives: We aimed to develop and externally validate a novel machine learning model that can classify CT image findings as positive or negative for SARS-CoV-2 reverse transcription polymerase chain reaction (RT-PCR).Methods: We used 3128 images from a wide variety of two-gate data sources for the development and ablation study of the machine learning model. A total of 633 COVID-19 cases and 2295 non-COVID-19 cases were included in the study. We randomly divided cases into a development set and ablation set at a ratio of 8:2. For the ablation study, we used another dataset including 150 cases of interstitial pneumonia among non-COVID-19 images. For external validation, we used 893 images from 740 consecutive patients at 11 acute care hospitals suspected of having COVID-19 at the time of diagnosis. The dataset included 343 COVID-19 patients. The reference standard was RT-PCR.Result: In ablation study, using interstitial pneumonia images, the specificity of the model were 0.986 for usual interstitial pneumonia pattern, 0.820 for non-specific interstitial pneumonia pattern, 0.400 for organizing pneumonia pattern. In the external validation study, the sensitivity and specificity of the model were 0.869 and 0.432, respectively, at the low-level cutoff, and 0.724 and 0.721, respectively, at the high-level cutoff.Conclusions: Our machine learning model exhibited a high sensitivity in external validation datasets and may assist physicians to rule out COVID-19 diagnosis in a timely manner. Further studies are warranted to improve model specificity.