AUTHOR=Mu Junhao , Kuang Kaiming , Ao Min , Li Weiyi , Dai Haiyun , Ouyang Zubin , Li Jingyu , Huang Jing , Guo Shuliang , Yang Jiancheng , Yang Li 

TITLE=Deep learning predicts malignancy and metastasis of solid pulmonary nodules from CT scans

JOURNAL=Frontiers in Medicine

VOLUME=Volume 10 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2023.1145846

DOI=10.3389/fmed.2023.1145846

ISSN=2296-858X

ABSTRACT=In clinic it’s difficult to distinguish malignancy and aggressivity of solid pulmonary nodules (PNs), incorrect assessments may lead to delayed diagnosis and increase the risk of complications. We developed and validated a deep learning-based model for the prediction of malignancy and local or distant metastasis in solid PNs based on CT images of the primary lesions at initial diagnosis. In this study We reviewed the data from consecutive patients with solid PNs at our institution from Jan 1st 2019 to April 30th 2022. The patients were divided into three groups: benign, Ia stage lung cancer and T1 stage lung cancer with metastasis. Each cohort was divided into training and testing group. Deep Learning System predicted malignancy and metastasis status of solid PNs based on CT images, then we compared malignancy prediction results among four different level clinicians. Experiments confirmed human-computer collaboration can further enhance diagnosis accuracy. We made a held-out testing set of 134 cases, 689 cases in total, our convolutional neural network model reached Area Under the ROC(AUC) of 80.37% on malignancy prediction and an AUC of 86.44% on metastasis prediction. In observer studies involving 4 clinicians, the proposed deep learning method outperformed a junior respiratory clinician and a 5-year respiratory clinician with considerable margins, was on par with a senior respiratory clinician, and only slightly inferior to a senior radiologist. Our human-computer collaboration experiment showed that by simply adding binary human diagnosis into model prediction probabilities, model AUC scores are improved to 81.80%-88.70% when combined with 3 out of 4 clinicians. In summary, Deep Learning method can accurately diagnose malignancy of solid PNs, and improve its performance when collaborating with human experts, and can be used to predict local or distant metastasis in patients with T1 stage lung cancer and facilitate the application of precision medicine.