Association of artificial intelligence-screened interstitial lung disease with radiation pneumonitis in locally advanced non-small cell lung cancer.
Authors
Affiliations (9)
Affiliations (9)
- Department of Radiation Oncology, University of Toronto, Toronto, Ontario, Canada.
- Techna Institute, University Health Network, Toronto, Ontario, Canada.
- Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada.
- Department of Biostatistics, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada. Electronic address: [email protected].
- Department of Radiation Oncology, University of Toronto, Toronto, Ontario, Canada; Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada.
- Division of Medical Oncology and Hematology, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada.
- Joint Department of Medical Imaging, University Health Network, Toronto, Ontario, Canada.
- Department of Medical Biophysics, University of Toronto, Toronto, Ontario Canada; Vector Institute for Artificial Intelligence, Toronto, Ontario, Canada; Peter Munk Cardiac Centre, University Health Network, Toronto, Ontario, Canada.
- Department of Radiation Oncology, University of Toronto, Toronto, Ontario, Canada; Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada. Electronic address: [email protected].
Abstract
Interstitial lung disease (ILD) has been correlated with an increased risk for radiation pneumonitis (RP) following lung SBRT, but the degree to which locally advanced NSCLC (LA-NSCLC) patients are affected has yet to be quantified. An algorithm to identify patients at high risk for RP may help clinicians mitigate risk. All LA-NSCLC patients treated with definitive radiotherapy at our institution from 2006 to 2021 were retrospectively assessed. A convolutional neural network was previously developed to identify patients with radiographic ILD using planning computed tomography (CT) images. All screen-positive (AI-ILD + ) patients were reviewed by a thoracic radiologist to identify true radiographic ILD (r-ILD). The association between the algorithm output, clinical and dosimetric variables, and the outcomes of grade ≥ 3 RP and mortality were assessed using univariate (UVA) and multivariable (MVA) logistic regression, and Kaplan-Meier survival analysis. 698 patients were included in the analysis. Grade (G) 0-5 RP was reported in 51 %, 27 %, 17 %, 4.4 %, 0.14 % and 0.57 % of patients, respectively. Overall, 23 % of patients were classified as AI-ILD + . On MVA, only AI-ILD status (OR 2.15, p = 0.03) and AI-ILD score (OR 35.27, p < 0.01) were significant predictors of G3 + RP. Median OS was 3.6 years in AI-ILD- patients and 2.3 years in AI-ILD + patients (NS). Patients with r-ILD had significantly higher rates of severe toxicities, with G3 + RP 25 % and G5 RP 7 %. R-ILD was associated with an increased risk for G3 + RP on MVA (OR 5.42, p < 0.01). Our AI-ILD algorithm detects patients with significantly increased risk for G3 + RP.