Objective To investigate the value of a deep learning-based dual-source computed tomography (DSCT) imaging model in differentiating benign and malignant pulmonary nodules. Methods The clinical data of 140 patients with pulmonary nodules admitted to a hospital from January 2023 to January 2025 were retrospectively analyzed. A total of 215 pulmonary nodules annotated on DSCT images were included as the study subjects and divided into a benign group (n=124) and a malignant group (n=91) based on the nature of the nodules. Another 60 patients with pulmonary nodules admitted to the same hospital from February 2025 to January 2026 were selected as the validation set, with 96 pulmonary nodules annotated on DSCT images. The chest DSCT images of the patients with pulmonary nodules were analyzed , and three convolutional neural network models (ResNet50, DenseNet121, ShuffleNetV2) were constructed using a PyTorch-based deep learning framework. All models employed the weights pre-trained on the ImageNet dataset for transfer learning. The diagnostic efficacy of the models was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, positive predictive value, and Youden index, and validated in the validation set. In the validation set, the Delong test was used to compare the diagnostic performance of the optimal deep learning model with that of three radiologists of different seniority. Results The malignancy rates of pulmonary nodules in the training set and validation set were 42.33% and 41.67%, respectively. The DenseNet121 model demonstrated superior AUC values in the training set and validation set (0.940, 95% CI=0.910-0.969; 0.909, 95%CI=0.847-0.971) compared to other models. The DenseNet121 model also showed favorable sensitivity (81.30%; 90.00%), specificity (90.30%; 76.80%), positive predictive value (86.80%; 91.50%), and Youden index (0.716; 0.668), along with good calibration and clinical benefits. The AUC of the DenseNet121 model was significantly higher than that of junior and intermediate-seniority radiologists (Z=2.637, P=0.008; Z=2.130, P=0.033). However, there was no statistically significant difference in AUC between the DenseNet121 model and the senior radiologists (Z=1.296, P=0.195). Conclusions The deep learning-based DSCT imaging model has high diagnostic value in differentiating benign and malignant pulmonary nodules and is expected to be an effective tool to assist junior and intermediate-seniority radiologists in diagnosis.
Key words
deep learning /
dual-source computed tomography /
pulmonary nodules /
benign-malignant differentiation /
diagnostic value
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
References
[1] Wang C, Chen B, Liang S,et al. China protocol for early screening, precise diagnosis, and individualized treatment of lung cancer[J]. Signal Transduct Target Ther, 2025, 10(1): 175.
[2] 于洪涛,王 锋,王 伟,等. 双源CT Flash扫描模式联合低剂量肺动脉CTA在肺栓塞诊断中的应用价值[J]. 医学影像学杂志,2025,35 (2):52-54,59.
[3] 刘敬伟,张西宁. 肺结节诊断方法的最新进展[J]. 中华外科杂志,2022,60 (5):498-503.
[4] 刘立兵,傅励瑶. 深度学习技术在医学影像分析中的应用与展望[J]. 新一代信息技术,2024,7 (1):24-28.
[5] 王 霄,朱恩照,艾自胜. 卷积神经网络的原理及其在医学影像诊断中的应用[J]. 中国医学物理学杂志,2022,39 (12):1485-1489.
[6] 中华医学会呼吸病学分会间质性肺疾病学组,中国医师协会呼吸医师分会间质性肺疾病工作委员会. 中国肺结节病诊断和治疗专家共识[J]. 中华结核和呼吸杂志,2019,42 (9):685-693.
[7] 冯国双. 诊断试验评价的样本量估算[J]. 慢性病学杂志,2022,23 (11):1657-1660.
[8] 程保国,李卫星. 双源CT双能量碘图成像对孤立性肺结节诊断价值分析[J]. 社区医学杂志,2022,20 (18):1027-1031.
[9] Isensee F, Jaeger P F, Kohl S A A,et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation[J]. Nat Methods, 2021,18(2): 203-211.
[10] Zhang T, Bur A M, Kraft S,et al. Gender, smoking history, and age prediction from laryngeal images[J]. J Imaging, 2023, 9(6): 109.
[11] 阿依古丽·托合提,班允清. 深度学习图像重建算法在CT成像中的应用进展[J]. 影像研究与医学应用,2024,8 (24):4-6.
[12] 黄梅萍,兰长青,王 洁.双源CT电子密度/有效原子序数(Rho/Z)在孤立性肺结节中的诊断价值[J]. 现代医用影像学,2024,33 (1):43-47.
[13] 王 钊,叶 彬,石昶霖. 64排双源炫速CT不同螺距肺动脉成像的图像质量对比[J]. 影像研究与医学应用,2025,9 (11):80-82.
[14] 黄志强,钟士江.机器学习在抑郁症辅助诊断中的应用研究进展[J].武警医学,2024,35(9):806-812.
[15] 陈宇辰,张昊坤,单婉婷,等. 基于舌象图像深度特征与支持向量机融合的肺结节风险分层预测研究[J]. 中医药信息,2025,42 (8):34-39.
[16] Saied M, Raafat M, Yehia S,et al. Efficient pulmonary nodules classification using radiomics and different artificial intelligence strategies[J]. Insights Imaging, 2023, 14(1): 91.
[17] 赵乘蔚,张超群,秦唯栋,等. 基于改进的ResNet-50深度学习的肺结节检测及其云端系统开发[J]. 现代计算机,2024,30 (21):187-193,205.
[18] Patharia P, Sethy P K, Nanthaamornphong A. Advancements and challenges in the image-based diagnosis of lung and colon cancer: a comprehensive review[J]. Cancer Inform, 2024, 23: 11769351241290608.
[19] Sailunaz K, Özyer T, Rokne J,et al. A survey of machine learning-based methods for COVID-19 medical image analysis[J]. Med Biol Eng Comput, 2023, 61(6): 1257-1297.
[20] Park J, Kim S, Lim J H,et al. Development of a multi-modal learning-based lymph node metastasis prediction model for lung cancer[J]. Clin Imaging, 2024, 114: 110254.
[21] 李文秀,罗 涛,张文静,等. 人工智能辅助诊断技术在超声心动图诊断房间隔缺损中的应用研究[J]. 中国医药,2022,17(2):274-277.
[22] Kumar Y, Shrivastav S, Garg K, et al. Automating cancer diagnosis using advanced deep learning techniques for multi-cancer image classification[J].Sci Rep,2024,14(1):25006.
[23] Liang C, Zheng M, Zou H,et al. Deep learning-based image analysis predicts PD-L1 status from 18F-FDG PET/CT images in non-small-cell lung cancer[J]. Front Oncol, 2024, 14: 1402994.
[24] Huo J, Min X, Luo T,et al. Computed tomography-based 3D convolutional neural network deep learning model for predicting micropapillary or solid growth pattern of invasive lung adenocarcinoma[J]. Radiol Med, 2024, 129(5): 776-784.
[25] Tanaka K, Nakada T A, Takahashi N, et al. Superiority of supervised machine learning on reading chest X-Rays in intensive care units[J].Front Med (Lausanne),2021,8:676277.
PDF(1997 KB)
