Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
MRI Deep Learning for Fully Automated Prediction of Occult Cervical Lymph Node Metastasis in Early-Stage Oral Tongue Squamous Cell Carcinoma: A Multi-Vendor Study
29 Pages Posted: 17 Sep 2024
More...Abstract
Background: Identifying cervical lymph node metastases (LNM) in early-stage oral tongue squamous cell carcinoma (OTSCC) is critical for predicting outcomes and selecting the most appropriate treatment strategy. This study aimed to develop and validate a fully automated magnetic resonance imaging (MRI) deep learning (DL) framework for tumor detection and prediction of occult LNM in early-stage OTSCC.
Methods: This retrospective study involved 348 patients from two centers with early-stage OTSCC. Patients from Center 1 were divided into a training cohort (n=163), a validation cohort (n=65), and an internal testing cohort (n=84) based on the vendors of MRI scanners. Patients from Center 2 were used as the external testing cohort (n=36). The You Only Look Once version 8 (YOLOv8) network was used as the backbone for automated tumor detection on conventional MRI. Two distinct image encoders—convolutional neural networks and vision transformer—were employed to construct the predictive model of LNM. The automated DL signature, which was the best DL model based on automated detection, was combined with significant clinical factors to develop the automated DL nomogram. Predictive performance was assessed using the area under the curve (AUC) and compared using the Delong’s test.
Findings: The YOLOv8 model demonstrated effective tumor detection ability on MRI, achieving mean average precisions (mAPs) of 0.877–0.973 for T1-stage tumors and mAPs of 0.689–0.961 for T2-stage tumors across the validation, internal, and external testing cohorts. The automated DL nomogram, which combined the automated DL signature and clinical T stage, achieved the best performance (AUCs: 0.871, 0.776, 0.909) and significantly outperformed clinical T stage alone (AUCs: 0.653, 0.611, 0.685, all P < 0.01) across the same cohorts.
Interpretation: The MRI-based DL framework can be used for fully automated prediction of LNM in early-stage OTSCC, potentially aiding in clinical decision-making.
Funding: This work was supported by funds from the National Scientific Foundation of China (No.82101992 and No.82172051) and Shanghai Ninth People’s Hospital (2022hbyjxys-rjl).
Declaration of Interest: The authors declare that they have no conflict of interest.
Ethical Approval: This retrospective study was approved by the Ethics Review Board of Shanghai Ninth People’s Hospital (IRB approval number: SH9H-2021-TK328-1) and First Affiliated Hospital of Guangxi Medical University (IRB approval number: 2024-S714-01); the requirement for written informed consent was waived.
Keywords: Oral tongue squamous cell carcinoma, magnetic resonance imaging, lymph node metastasis, deep learning
Suggested Citation: Suggested Citation