AUTHOR=Shen Jing , Wang Shaobin , Guan Hui , Di Mingyi , Liu Zhikai , Chen Qi , Li Mei , Shen Jie , Hu Ke , Zhang Fuquan 

TITLE=Artificial intelligence in automatic image segmentation system for exploring recurrence patterns in small cell carcinoma of the lung

JOURNAL=Frontiers in Oncology

VOLUME=Volume 15 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2025.1534740

DOI=10.3389/fonc.2025.1534740

ISSN=2234-943X

ABSTRACT=BackgroundThe integration of artificial intelligence (AI) in automatic image segmentation systems offers a novel approach to evaluating the clinical target volume (CTV) in small cell lung cancer (SCLC) patients. Utilizing imaging recurrence data, this study applies a recursive feature elimination algorithm to model and predict patient prognoses, aiming to enhance clinical guidance and prediction accuracy.Materials and MethodsThis research analyzed data from SCLC patients who received curative radiotherapy from January 1, 2010, to December 30, 2021, and had comprehensive follow-up records including pre- and post-treatment imaging. An AI-driven image segmentation system segmented the initial CTV, evaluating 110 clinical parameters. The recursive feature elimination method selected pertinent features, and a random forest-based recursive prediction model was developed to establish a clinically viable recurrence prediction model.Results1. Local Control Analysis: A study of 180 patients, with a median follow-up duration of 36 months, revealed that 94 experienced recurrences, while 86 did not. Factors influencing local control rates included gender (male), age (>60 years), T stage, smoking index, and tumor size. Notably, tumor size (≥ 5cm) emerged as an independent factor significantly impacting local control rates, with a Hazard Ratio (HR) of 1.635 (95% CI: 1.055-2.536, p=0.028). 2. Recurrence Analysis: Tumor size (≥ 5cm) was also closely linked to patient local control rates, with a 3-year Local Control Rate Failure (LCRF) contrasting sharply between larger tumors (61.1%) and smaller tumors (86.7%, p=0.004). Upon analyzing recurrence patterns among 94 patients, a total of 170 instances were examined. Recurrence was most prevalent in regions 10R, 10L, 4R, and 7, accounting for 67.65% (115/170) of cases, while regions 2L and 3P showed no recurrences. The initial region of the primary tumor or metastatic lymph nodes was identified as a critical recurrence area, with a 100% recurrence rate in patients whose initial tumor region included 10 specific regions. The recurrence rates for initial tumor regions involving 4R, 7, 11R, and 11L ranged between 41.6% and 45.5%. 3. Development of a recurrence prediction model: utilizing an AI-powered automatic image segmentation system, multidimensional partition parameters, including 110 clinical variables, were analyzed. The recursive feature elimination method facilitated efficient feature selection. From this, a recurrence prediction model for small cell lung cancer was developed using a random forest algorithm, achieving a clinically significant accuracy rate of 77%. This model provides a reliable basis for enhancing the clinical application and decision-making process for medical practitioners.ConclusionThe utilization of AI-based automatic image segmentation system for delineating the initial CTV has proven pivotal. Analysis and modeling of recurrent images reveal that the initial GTV and GTVnd are critical regions for recurrence. Leveraging partition parameter and variable information, we constructed a clinically viable recurrence prediction model. This model significantly aids in guiding the precise clinical targeting of treatment areas, demonstrating the potential of AI to enhance patient management and treatment outcomes in small cell lung cancer.