Belfiore Maria Paola, Sansone Mario, Ciani Giovanni, Patanè Vittorio, Genco Carlotta, Grassi Roberta, Savarese Giovanni, Montella Marco, Monti Riccardo, Cappabianca Salvatore, Reginelli Alfonso
Department of Precision Medicine, University of Campania Luigi Vanvitelli, Naples, Italy.
Department of Electrical Engineering and Information Technology, University of Naples "Federico II", Naples, Italy.
Thorac Cancer. 2025 Jul;16(13):e70115. doi: 10.1111/1759-7714.70115.
Nonsmall cell lung cancer (NSCLC) remains a significant global health burden, necessitating advancements in diagnostic and prognostic strategies. Liquid biopsy and radiomics offer promising avenues for enhancing preoperative assessment and treatment planning in NSCLC.
This prospective study enrolled 60 NSCLC patients who underwent both computed tomography (CT)-guided biopsy and liquid biopsy. Radiomic features were extracted from CT images, and circulating tumor DNA (ctDNA) was sequenced to identify genetic mutations. Machine learning algorithms were employed to assess the association between radiomic features and gene mutations.
Among 57 patients with available data, associations between radiomic features and gene pairs mutation obtained from liquid biopsy exhibited moderate accuracy (approximately 0.60), with texture features demonstrating higher importance. However, when predicting the combined mutation status of gene pairs (e.g., EGFR and ROS1), the classification task involved three classes and yielded substantially lower accuracy (approximately 0.30), likely due to class imbalance and increased complexity.
Our findings demonstrate a moderate association between radiomic features and single gene mutations detected through liquid biopsy in NSCLC patients, with classification accuracies reaching approximately 0.60. In contrast, classification performance significantly declined (to ~0.30) when gene mutation pairs were used as targets, likely due to increased complexity and class imbalance. Notably, second-order texture features showed the highest importance in the models. These preliminary results suggest that radiomics may capture aspects of tumor biology reflected in liquid biopsy, warranting further validation in larger, well-balanced cohorts.
The integration of liquid biopsy and radiomics holds promise for enhancing preoperative assessment and personalized treatment strategies in NSCLC. Further research on larger cohorts is warranted to validate the findings and translate them into clinical practice.
University of Campania Trial Board UC20201112-24997.
非小细胞肺癌（NSCLC）仍然是一项重大的全球健康负担，这使得诊断和预后策略需要取得进展。液体活检和放射组学为加强NSCLC的术前评估和治疗规划提供了有前景的途径。
这项前瞻性研究纳入了60例接受计算机断层扫描（CT）引导活检和液体活检的NSCLC患者。从CT图像中提取放射组学特征，并对循环肿瘤DNA（ctDNA）进行测序以识别基因突变。采用机器学习算法评估放射组学特征与基因突变之间的关联。
在57例有可用数据的患者中，放射组学特征与通过液体活检获得的基因对突变之间的关联显示出中等准确性（约为0.60），其中纹理特征显示出更高的重要性。然而，在预测基因对（如EGFR和ROS1）的联合突变状态时，分类任务涉及三个类别，准确性大幅降低（约为0.30），这可能是由于类别不平衡和复杂性增加所致。
我们的研究结果表明，NSCLC患者中放射组学特征与通过液体活检检测到的单基因突变之间存在中等关联，分类准确率约为0.60。相比之下，当将基因突变对用作目标时，分类性能显著下降（至约0.30），这可能是由于复杂性增加和类别不平衡所致。值得注意的是，二阶纹理特征在模型中显示出最高的重要性。这些初步结果表明，放射组学可能捕捉到液体活检中反映的肿瘤生物学方面，需要在更大的、平衡良好的队列中进行进一步验证。
液体活检和放射组学的整合有望加强NSCLC的术前评估和个性化治疗策略。有必要对更大的队列进行进一步研究，以验证研究结果并将其转化为临床实践。
坎帕尼亚大学试验委员会UC20201112 - 24997。
