Artificial intelligence for accurate tumor size assessment and non-invasive adenocarcinoma prediction in small-sized...

Accurate preoperative imaging is essential for improving the treatment of small lung cancers. Precise identification of non-invasive adenocarcinomas is critical for determining the suitability of sublobar resection. Conventional methodologies frequently demonstrate variability, particularly for small tumors or ground-glass nodules (GGNs). Artificial intelligence (AI) offers a consistent and objective alternative, enhancing non-invasive cancer diagnosis and facilitating more effective treatment decisions.

A retrospective analysis was conducted on 324 patients who underwent surgical resection for small-sized lung adenocarcinomas at Tokyo Medical University. The Synapse Vincent system (Fujifilm Corporation, Japan) was employed to measure tumor size and classify the nodules as GGN (AI-GGN) or non-GGN (non-AI-GGN) based on confidence scores. The ability of AI to predict pathological non-invasive adenocarcinomas was evaluated.

AI-measured tumor sizes were significantly more accurate than those measured by thoracic surgeons (p < 0.001) AI-GGN demonstrated a high specificity of 98.3% for predicting pathological non-invasive adenocarcinoma, closely aligned with the 98.3% specificity of the traditional consolidation tumor ratio (CTR) method. The positive predictive values of AI-GGN and CTR were similarly high, (98.5% and 98.2%, respectively), confirming the effectiveness of both methods in identifying non-invasive adenocarcinomas.

AI technology significantly enhances the precision of tumor size measurement and identification of non-invasive adenocarcinomas in small-sized lung tumors. By providing objective and automated evaluations, AI can refine surgical planning and decision-making. Further prospective multicenter studies are warranted to validate these findings and to fully integrate AI into clinical practice, ultimately improving patient outcomes.