Hitting the Bull’s AI: Artificial Intelligence-derived Imaging Features and their Association with Outcomes in CT-guided Lung Biopsy, a Retrospective Study

Abstract

PURPOSE: This study aims to evaluate whether quantitative imaging features analyzed by an artificial intelligence (AI) tool are associated with success rate, histopathological results, and complication risks of CT-guided lung biopsies. METHODS: A retrospective study was conducted on 120 CT-guided biopsies of suspicious pulmonary lesions with pathology reports. Associations between technical success, histopathology, occurrence of peri-interventional complications, as well as intervention-related factors such as lesion diameter and biopsy pathway and the AI-derived parameters lesion volume, malignancy probability and emphysema ratio were assessed using t-test for continuous data and Chi-square test for categorical data. Adjusted multivariate logistic regression models and predictive performance of AI parameters were calculated. RESULTS: Ninety-eight of 120 biopsies (81.7 %) were technically successful. Peri-interventional pneumothorax occurred in 65 % of cases, 26.7 % needed drainage. Alveolar hemorrhage was documented in 53.3 %, high-grade hemorrhage in 30.8 %. Adjusted regression models showed significant association of AI-derived lesion volume with technical success (OR = 1.30, CI 1.00; 1.69), AI malignancy chance with histopathologically confirmed malignancy (OR = 1.17, CI 1.08; 1.28) and AI emphysema ratio with increased risk of pneumothorax requiring chest tube insertion (OR = 1.29, CI 1.12; 1.48). For alveolar hemorrhage, only AI lesion volume (p = 0.011) showed a significant inverse correlation in unadjusted models, while adjusted models identified emphysema ratio as the relevant AI feature. CONCLUSIONS: AI derived imaging features show significant association with complication risks in CT-guided lung biopsies, like pneumothorax and alveolar hemorrhage. This may allow stricter patient selection and better execution of biopsies, which may lead to improved outcomes and patient safety. Additionally, software's high association with histopathological malignancy supports reconsideration of its role in guiding the indication for lung biopsy in the assessment of pulmonary lesions. CLINICAL RELEVANCE STATEMENT: With the advancing integration of AI tools into radiological workflows, this study highlights AI's potential to provide pre-interventional risk stratification and outcome predictions for CT-guided lung biopsies, which are the gold standard for sampling peripheral lung lesions.

Ähnliche Arbeiten