Tumor radiomics signature for artificial neural network-assisted detection of neck metastasis in patient with tongue cancer - 11/08/21

Doi : 10.1016/j.neurad.2021.07.006

Yi-Wei Zhong ^a,^{^{1
Yi-Wei Zhong, M.M., and Yin Jiang, Ph.D. contributed equally to this work.}}, Yin Jiang ^b,^c,^{^{1
Yi-Wei Zhong, M.M., and Yin Jiang, Ph.D. contributed equally to this work.}}, Shuang Dong ^a, Wen-Jie Wu ^a,^⁎ , Ling-Xiao Wang ^d,^e, Jie Zhang ^a, Ming-Wei Huang ^a
^a Department of Oral and Maxillofacial Surgery, Peking University School and Hospital of Stomatology, Beijing, PR China
^b Department of Physics, Beihang University, Beijing, PR China
^c Beijing Advanced Innovation Center for Big Data-based Precision Medicine, Beihang University, Beijing, PR China
^d Department of Physics, Tsinghua University, Beijing, PR China
^e Frankfurt Institute for Advanced Studies, Frankfurt am Main, Germany

^⁎Corresponding author at: No. 22 Zhongguancun South Avenue, Department of Oral and Maxillofacial Surgery, Peking University School and Hospital of Stomatology, Beijing, 100081, China.Department of Oral and Maxillofacial SurgeryPeking University School and Hospital of StomatologyNo. 22 Zhongguancun South AvenueBeijing100081China

Sous presse. Épreuves corrigées par l'auteur. Disponible en ligne depuis le Wednesday 11 August 2021
Cet article a été publié dans un numéro de la revue, cliquez ici pour y accéder

Abstract

Background and purpose

To determine the neck management of tongue cancer, this study attempted to construct an artificial neural network (ANN)-assisted model based on computed tomography (CT) radiomics of primary tumors to predict neck lymph node (LN) status in patients with tongue squamous cell carcinoma (SCC).

Materials and methods

Three hundred thirteen patients with tongue SCC were retrospectively included and randomly divided into training (60%), validation (20%) and internally independent test (20%) sets. In total, 1673 feature values were extracted after the semiautomatic segmentation of primary tumors and set as input layers of a classical 3-layer ANN incorporated with or without clinical LN (cN) status after dimension reduction. The receiver operating characteristic (ROC) curve, accuracy (ACC), sensitivity (SEN), specificity (SPE), area under curve (AUC) and Net Reclassification Index (NRI), were used to evaluate and compare the models.

Results

Four models with different settings were constructed. The ACC, SEN, SPE and AUC reached 84.1%, 93.1%, 76.5% and 0.943 (95% confidence interval: 0.891-0.996, p<.001), respectively, in the test set. The NRI of models compared with radiologists reached 40% (p<.001). The occult nodal metastasis rate was reduced from 30.9% to a minimum of 12.7% in the T1-2 group.

Conclusion

ANN-based models that incorporated CT radiomics of primary tumors with traditional LN evaluation were constructed and validated to more precisely predict neck LN metastasis in patients with tongue SCC than with naked eyes, especially in early-stage cancer.

Le texte complet de cet article est disponible en PDF.