Development of imaging protocol and radiomics-based nomogram for assessing lesion reversibility in connective tissue disease-associated interstitial lung disease

Assessing the reversibility of connective tissue disease associated interstitial lung disease (CTD-ILD) is critical for treatment decision-making, yet standardized imaging criteria and baseline predictive tools remain limited. We aimed to establish a CT-based protocol to determine lesion reversibility and develop a combined radiomics–morphology nomogram to predict baseline reversibility. We retrospectively analyzed consecutive CTD-ILD patients (January 2016–June 2022, diagnosed via ACR/EULAR criteria) from three branches of Renji Hospital. CTD-ILD diagnoses were established by a multidisciplinary team of respiratory physicians and thoracic radiologists based on clinical manifestations and chest CT imaging. Patients from the South Branch were randomly split into training and internal validation cohorts (7:3), while patients from the East and West Branches formed an independent external validation cohort. Lesion reversibility was adjudicated using longitudinal CT series across six lung zones: the scan with the greatest lesion extent was defined as baseline (CT0); completely reversible (CR) lesions were defined as complete absorption of baseline lesions on any follow-up CT, whereas non-CR required persistence of baseline lesions on both 3–6-month (CTF1) and 12-month (CTF2) follow-up CTs. Zonal-level radiomics signatures and visual CT features were used to develop radiomics, visual, and combined nomogram models. The final cohort included 153 patients (median age, 53 years; interquartile range, 44–62 years), with 40 males (26.1%) and 113 females (73.9%) with 575 affected lung zones. Cohorts comprised 66/28/59 patients (training/internal validation/external validation). The radiomics model showed moderate discrimination (areas under the curve [AUCs] 0.72/0.77/0.73), while the visual model performed similarly or better (AUCs 0.82/0.87/0.78). The combined nomogram achieved the best overall performance (AUCs 0.86/0.90/0.82), with significant improvement over the visual model in the training and external validation cohorts. We established a longitudinal imaging protocol to standardize reversibility assessment and validated a combined nomogram that integrates quantitative and visual data. This tool accurately predicts lesion reversibility at baseline, potentially facilitating personalized treatment strategies for CTD-ILD patients.