Combined Radiomic and Visual Assessment for Improved Detection of Lung Adenocarcinoma Invasiveness on Computed...

Area of Science:

• Pulmonary Medicine
• Radiology
• Oncology

Background:

• Distinguishing between adenocarcinoma-in-situ (AIS), minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (INV) is crucial for surgical intervention timing and approach.
• Small, ground-glass lesions on computed tomography (CT) scans present a diagnostic challenge in accurately determining invasiveness.

Purpose of the Study:

• To develop and evaluate a quantitative imaging method using radiomics to determine the invasiveness of small, ground-glass lung lesions on CT scans.
• To differentiate between AIS, MIA, and invasive adenocarcinoma (INV) based on CT-derived radiomic features.

Main Methods:

• A total of 268 patients with resected semisolid lesions (<=3 cm) and confirmed histopathology of MIA/AIS or INV were included from four institutions.
• 248 radiomic texture features (intratumoral and peritumoral) were extracted from manually annotated lung nodules on CT scans.
• A machine learning classifier was trained on 40% of the data and tested on the remaining 60% to differentiate lesion invasiveness.

Main Results:

• The top five radiomic features (four intratumoral, one peritumoral) differentiated INV from MIA/AIS nodules with an AUC of 0.917 (training) and 0.863 (testing).
• The radiomics model showed high accuracy in differentiating invasive adenocarcinoma from MIA cases across different size categories (AUCs ranging from 0.76 to 0.95).
• An integrated model combining the radiomics classifier with radiologists' scores achieved the best performance on the test set (AUC=0.909).

Conclusions:

• Advanced image analysis via radiomics, when added to routine visual assessment of CT scans, can improve the differentiation of lung adenocarcinoma subtypes.
• This quantitative imaging approach can aid in clinical decision-making for managing small, semisolid lung nodules.
• Further prospective validation is warranted to confirm the utility of radiomics in clinical practice.