Deep learning-based solid component measuring enabled interpretable prediction of tumor invasiveness for lung...

Area of Science:

• Pulmonary medicine
• Radiology
• Artificial Intelligence in Medicine

Background:

• The solid component of subsolid nodules (SSNs) is crucial for assessing pathological invasiveness.
• Preoperative assessment of the solid component in SSNs currently lacks a standardized reference.

Purpose of the Study:

• To develop and evaluate an AI algorithm for measuring the solid component ratio in SSNs.
• To compare the diagnostic performance and efficiency of the AI algorithm against manual radiologist measurements for predicting lung adenocarcinoma invasiveness.

Main Methods:

• A retrospective study involving 379 patients (278 primary, 101 validation).
• An AI algorithm was developed to measure the 1D, 2D, and 3D solid component ratios in SSNs.
• Radiologists manually measured the consolidation to tumor ratio (CTR) twice, four weeks apart. Diagnostic performance was assessed using the area under the receiver-operating characteristic curve (AUC).

Main Results:

• The AI algorithm demonstrated superior predictive performance in measuring the 3D solid component ratio (AUC: 0.811) compared to manual methods (AUC: 0.697) in the primary dataset.
• The AI 3D method showed comparable or superior diagnostic performance to 1D and 2D measurements and outperformed radiologist measurements in the validation dataset (AUC: 0.803 vs 0.682).
• AI measurements were significantly faster (approximately 60 times) and showed better consistency than manual measurements, which had 7.9% cases with poor consistency over four weeks.

Conclusions:

• AI-based 3D measurement of solid components in SSNs is an effective and objective method for predicting pathological invasiveness.
• This AI approach can serve as a valuable preoperative indicator for pathological invasiveness in lung adenocarcinoma.