Prediction of 131I Uptake in Lung Metastases of Differentiated Thyroid Cancer Using Deep Learning

Abstract Objective: An accurate assessment the 131I accumulation capacity in lung metastases of differentiated thyroid cancer (DTC) is pivotal to inform radioiodine therapy and avoid invalid 131I administration. Thus, to develop a deep convolutional neural network (DCNN) model for the prediction of the 131I uptake in lung metastases of DTC before radioiodine therapy.Methods: In this retrospective, multicentre, population-based cohort study, we collected chest CT images datasets for DTC patients with lung metastases from three hospitals in China. Pulmonary metastases were classified into two categories based on the post-therapeutic 131I whole-body scan: 131I-avid (positive 131I uptake) and non-131I-avid (negative 131I uptake). For DCNN model development, patients were assigned to the primary dataset (140 patients with 131I-avid, 121 with non-131I-avid ). For model validation, patients were assigned to the internal validation dataset (36 patients with 131I-avid, 23 with non-131I-avid), external validation datasets 1 (25 patients with 131I-avid, 18 with non-131I-avid), and 2 (23 patients with 131I-avid, 18 with non-131I-avid). Using these datasets, we assessed the performance of our model ResNeSt50 and compared with two models as Inception V3 and ResNet50.Results: Compared to Inception V3 and ResNet50, our model ResNeSt50 demonstrated the highest prediction performance in internal (AUC = 0.722, 95%CI 0.716–0.725), external 1 (AUC = 0.720, 95%CI 0.691–0.749), and external 2 (AUC = 0.731 95%CI 0.713, 0.748) validation datasets.Conclusions: We developed a simple and robust DCNN model for predicting the 131I uptake in lung metastases of DTC before radioiodine therapy, which can provide improved screening for patients who can benefit from 131I therapy.Trial registration Chinese Clinical Trial Registry (ChiCTR), ChiCTR1800018047. Registered on 28 August 2018.