Assessing Interstitial Lung Abnormalities With a DL-IQ-UIP Classifier and LTA Algorithms: Links to Mortality, Cancer, All-cause and Pneumonia Hospitalization

Le résumé fourni par la source

Abstract Rationale: The widespread adoption of low-dose-CT (LDCT) for lung cancer screening has created opportunities to identify undiagnosed interstitial lung disease (ILD) through finding interstitial lung abnormalities (ILA). ILA are frequently present, but often under-recognized, in these high-risk populations. ILA are associated with adverse clinical outcomes, including mortality and progression to ILD, underscoring the need for timely diagnosis. We investigated the association between a deep learning (DL)-based IQ-UIP classifier and Lung Texture Analysis (LTA) (4DMedical, Los Angeles) quantitative measurements for ILA and key outcomes: mortality; lung cancer incidence; and all-cause and pneumonia-related hospitalizations. Methods: This multicenter, retrospective cohort study included patients undergoing LDCT at Lahey Hospital and Medical Center (LHMC) (2012-2017) and Mt. Auburn Hospital (MAH) (2015-2017) per NCCN high-risk criteria for lung cancer screening. IQ-UIP and LTAwere utilized to generateIQUIP-high-riskand IQUIP-moderate-risk scores andhoneycombing and reticulation extent on LDCT. Follow-up through 2019 (LHMC) and 2020 (MAH) tracked key clinical outcomes. Cox proportional hazards models assessed associations between IQ-UIP and LTA for each outcome, adjusting for age, sex, BMI, smoking status/pack-years, with significance set at p<0.05 for associations that replicated in the MAH cohort. Results: Of 4673 scans at LHMC and 1271 at MAH, 4644 (99.4%) and 1253 (98.6%) were processed for IQ-UIP; and 3951 (84.5%) and 1253 (98.6%) for LTA, respectively. Mean age was 62.4 years (54.4% male) at LHMC and 64.3 years (49.5% male) at MAH. There were 11 and 5 IQ-high-risk; and 29 and 7 IQ-moderate-risk scans at LHMC and MAH, respectively. Hazard ratios (HRs) for mortality in LHMC and MAH cohorts were 11.0 and 12.54 for IQ-high-risk; 3.99 and 8.45 for IQ-moderate-risk; 4.37 and 1.73 for honeycombing, respectively. In LHMC cohort, IQ-high-risk (HR 6.72), IQ-moderate-risk (HR 3.61) and honeycombing (HR 5.30) were associated with lung cancer (MAH without association). Association with all-cause hospitalization was highest among LHMC IQ-high-risk (HR 5.90) and MAH IQ-moderate-risk patients (HR 4.32). Pneumonia-related hospitalizations showed the strongest associations: HRs in LHMC and MAH cohorts of 13.20 and 8.79 for IQ-high-risk; 4.99 and 6.99 for IQ-moderate-risk; 5.20 and 1.85 for honeycombing (Table#1). Conclusion: This study highlights the strong association between ILA identified by quantitative algorithms and pneumonia-related hospitalizations, suggesting that ILA are often misclassified as pneumonia, contributing to diagnostic delays in ILD. These results support the use of DL-based tools within lung screening programs to detect and classify ILA patterns, potentially enabling early diagnosis and appropriate intervention to reduce the clinical burden of undiagnosed ILD.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.