Artificial Intelligence-Advanced Imaging for Solid-Type Lung Adenocarcinoma: Towards Greater Clinical Relevance

Non-small-cell lung cancer (NSCLC) lesions appearing solid on imaging are highly likely to be malignant. However, "solid" has a subjective definition. This study aimed to determine whether artificial intelligence (AI)-based imaging analysis can offer a more objective and clinically meaningful definition of solid tumours, specifically in regards to lymph node metastasis and prognosis.

This study included 216 patients with cN0 lung adenocarcinoma (pathological invasive diameter ≤ 30 mm) who underwent lobectomy with lymph node dissection. AI software was used to calculate the consolidation-to-tumour diameter (cD/tD) and volume (cV/tV) ratios, which were then compared with radiologist-defined cD/tD. To determine the optimal cutoff values, correlations between the pathological invasive diameter/tumour diameter ratio (PathoiD/tD) and lymph node metastasis were evaluated. Subsequently, these values were applied to a subset of tumours measuring ≤20 mm (n = 117) to determine their potential use in candidate selection for limited resection.

A cV/tV cutoff value of ≥0.72 accurately predicted lymph node metastasis and improved concordance between AI and radiologist solid tumour assessment while maintaining a similar PathoiD/tD to the conventional (cD/tD = 1.0) threshold. Among tumours measuring ≤20 mm in the low solid-component group (cV/tV <0.72), lymph node metastases were not observed; the 5-year recurrence-free survival rate was 100%.

An AI-based volumetric analysis using a cV/tV threshold of ≥0.72 showed potential for predicting lymph node metastasis in this single-centre retrospective study of lung adenocarcinoma ≤30 mm. External validation in diverse, multicentre cohorts is essential before clinical implementation.