Developing and Validation of a Multimodal-Based Machine Learning Model for Diagnosis of Usual Interstitial Pneumonia: A Prospective Multicenter Study.

Abstract

Usual interstitial pneumonia (UIP) indicates a poor prognosis, and there is significant heterogeneity in the diagnosis of UIP, necessitating an auxiliary diagnostic tool. Can a machine learning (ML) classifier using radiomics features and clinical data accurately identify UIP from patients with interstitial lung disease (ILD)? This data from a prospective cohort includes 5,321 sets of high-resolution CT (HRCT) images from 2,901 patients with ILD (male, 63.5%; mean age ± SD, 61.7 ± 10.8 years) across 3 medical centers. Multimodal data, including whole-lung radiomics features on HRCT scan, demographics, smoking status, pulmonary function, and comorbidity data, were extracted. An XGBoost and logistic regression were used to design a nomogram predicting UIP or not. The area under the receiver operating characteristic curve (AUC) and Cox regression for all-cause mortality were used to assess the diagnostic performance and prognostic value of models, respectively. A total of 5,213 HRCT images were divided into the training group (n = 3,639), the internal testing group (n = 785), and the external validation group (n = 789). UIP prevalence was 43.7% across the whole data set, with 42.7% and 41.3% for the internal validation set and external validation set, respectively. The radiomics-based classifier had an AUC of 0.790 in the internal testing set and 0.786 for the external validation data set. Integrating multimodal data improved AUCs to 0.802 and 0.794, respectively. The performance of the integration model was comparable with a pulmonologist with > 10 years of experience in ILD. Within 522 patients deceased during a median follow-up period of 3.37 years, the multimodal-based ML model-predicted UIP pattern was associated with high all-cause mortality risk (hazard ratio, 2.52; P < .001). The classifier combining radiomics and clinical features showed strong diagnostic performance across varied UIP prevalence. This multimodal-based ML model could serve as an adjunct in the diagnosis of UIP. ClinicalTrials.gov; No.: NCT04370158; URL: www. gov.