Searchable abstracts of presentations at key conferences in endocrinology
Endocrine Abstracts (2023) 90 EP951 | DOI: 10.1530/endoabs.90.EP951

ECE2023 Eposter Presentations Thyroid (128 abstracts)

Application of a novel multimodal-based deep learning model for prediction of recurrence in patients with papillary thyroid carcinoma

Dong-Hwa Lee 1 , Jee-Woo Choi 2 , Geun-Hyeong Kim 2 , Seung Park 3 & Hyun Jeong Jeon 1


1Chungbuk National University College of Medicine and Chungbuk National University Hospital, Internal Medicine, Cheongju, South Korea; 2Chungbuk National University Hospital, Medical AI Research Team, Cheongju, South Korea; 3Biomedical Engineering, Chungbuk National University Hospital, Cheongju, South Korea


Background: Thyroid cancer is the most common endocrine malignancy, and papillary thyroid carcinoma (PTC) is the most frequent type of thyroid malignancy. Although the mortality rate is low, some patients experience cancer recurrence during follow-up periods. Early detection of recurrence helps improve the outcomes of patients and reduce the socioeconomic burden. In this study, we investigated the accuracy of a novel multi-modal prediction model by simultaneously analyzing numeric and time-series data to predict recurrence in patients with PTC after thyroidectomy.

Methods: We analyzed patients with thyroid carcinoma who underwent thyroidectomy at Chungbuk National University Hospital between January 2006 and December 2021. To detect PTC recurrence, we acquired clinical data, including demographic information, ultrasonography (US) reports, pathology reports, whole-body iodine scan, and thyroid function test results. We propose a novel multimodal-based deep learning model for predicting PTC recurrence. The proposed model uses numerical data, including clinical information at the surgery, and time-series data, including TFT results after the surgery. For the model training with unbalanced data, we employed weighted binary cross-entropy with weights of 0.8 for the positive (recurrence) group and 0.2 for the negative group.

Results: Our dataset consists of 1,613 patients, including patients who 1,550 non-recurrence PTC and 63 recurrence PTC, who underwent thyroidectomy. Patients with recurrence had larger tumor size, more multiplicity, and a higher frequent male ratio than those without recurrence. We performed 4-fold cross-validation on the dataset to evaluate the model performance. The proposed model achieved an average AUROC of 0.9622, F1-score of 0.4603, sensitivity of 0.9042, specificity of 0.9077. When applying our proposed model, experimental results show that it can predict recurrence at least one year before recurrence.

Conclusions: Our study results demonstrated that the multi-modal prediction model for predicting PTC recurrence after thyroidectomy showed good performance. In clinical practice, it might help in the early detection of recurrence during follow-up in patients with PTC after thyroidectomy.

Volume 90

25th European Congress of Endocrinology

Istanbul, Turkey
13 May 2023 - 16 May 2023

European Society of Endocrinology 

Browse other volumes

Article tools

My recent searches

No recent searches.