ChatGPT-assisted deep learning model for thyroid nodule analysis: beyond artifical intelligence
İsmail Meşe, Neslihan Gökmen Inan, Ozan Kocadağlı, Artür Salmaslıoğlu, Düzgün Yıldırım
Abstract
AIMS: To develop a deep learning model, with the aid of ChatGPT, for thyroid nodules, utilizing ultrasound images. The cytopathology of the fine needle aspiration biopsy (FNAB) serves as the baseline. MATERIAL AND METHODS: After securing IRB approval, a retrospective study was conducted, analyzing thyroid ultrasound images and FNAB results from 1,061 patients between January 2017 and January 2022. Detailed examinations of their demographic profiles, imaging characteristics, and cytological features were conducted. The images were used for training a deep learning model to identify various thyroid pathologies. ChatGPT assisted in developing this model by aiding in code writing, preprocessing, model optimization, and troubleshooting. RESULTS: The model demonstrated an accuracy of 0.81 on the testing set, within a 95% confidence interval of 0.76 to 0.87. It presented remarkable results across thyroid subgroups, particularly in the benign category, with high precision (0.78) and recall (0.96), yielding a balanced F1-score of 0.86. The malignant category also displayed high precision (0.82) and recall (0.92), with an F1-score of 0.87. CONCLUSIONS: The study demonstrates the potential of artificial intelligence, particularly ChatGPT, in aiding the creation of robust deep learning models for medical image analysis.