(P 273) RADIOMIC PREDICTORS OF SPONTANEOUS PNEUMOTHORAX DURING TYROSINE KINASE INHIBITOR THERAPY FOR LUNG METASTASES

Objective: Tyrosine kinase inhibitors (TKIs) are increasingly used to treat pulmonary metastatic disease in osteosarcoma (OS) and synovial sarcoma (SS). However, TKIs have been associated with spontaneous pneumothorax (sPNX).
This study aims to identify clinical and radiomic features associated with sPNX while on TKI therapy.

Methods: We performed a retrospective review of patients aged 15–39y with OS or SS lung metastases treated with TKI therapy between 2007–2023. Demographics, therapy and outcomes data were collected under IRB approval. The pretreatment chest CT was reviewed, and all lung lesions were segmented. Radiomic features were extracted from all segmented lung lesions using a well validated feature extraction package. The primary outcome was the occurrence of sPNX during TKI therapy and up to a month after. Univariate mixed-effects logistic regression was used to assess associations between individual features and sPNX, controlling the false discovery rate at 5%.

Results: A total of 58 patients were included (median age= 28; 64% male, 26 OS and 32 SS), of which, 43 (74%) patients had prior thoracic surgery. The median number of lesions per patient was 6 (range1-207), and 51(88%) patients had lesions abutting the pleura.
Seven patients (3 OS, 4 SS) experienced TKI related sPNX, including two with bilateral events, for a total of nine sPNX events. The number of lung lesions was significantly higher in patients with sPNX (median 17 vs. 5; p = 0.03). Other clinical factors including presence of pleural abutment and history of thoracic surgery were not significantly associated with sPNX.
The radiomics analysis included 963 features across 1121 lesions. Sixteen representative features were identified from the associated classes: Gray-Level Co-occurrence Matrix (GLCM), Gray-Tone Difference Matrix (GTDM), Grey Level Run Length (GLRL), Laplacian of Gaussian (LoG) and Intensity. Among these the GLRL class feature (OR 0.04, 95% CI: 0.01-0.34) had the strongest association suggesting a strong protective effect.

Conclusion: Our study identified clinical and radiomic features associated with sPTX during TKI therapy. These results suggest the potential for modeling risk based on pre-treatment cross-sectional imaging.