Intelligent Diagnosis of Thyroid Ultrasound Imaging Using an Ensemble of Deep Learning Methods

Corina Maria Vasile,Andreea Valentina Iacob,Andrei Ioan Drocaş,Anca Marilena Ungureanu,Dragoş Ovidiu Alexandru,Carmen Elena Niculescu,Cristian Gheonea,Alice Elena Ghenea,Mihaela Popescu,Anca Loredana Udriștoiu,Ștefan Udriștoiu,Lucian Gheorghe Gruionu,Gabriel Gruionu

doi:10.3390/medicina57040395

Abstract

Background and Objectives: At present, thyroid disorders have a great incidence in the worldwide population, so the development of alternative methods for improving the diagnosis process is necessary. Materials and Methods: For this purpose, we developed an ensemble method that fused two deep learning models, one based on convolutional neural network and the other based on transfer learning. For the first model, called 5-CNN, we developed an efficient end-to-end trained model with five convolutional layers, while for the second model, the pre-trained VGG-19 architecture was repurposed, optimized and trained. We trained and validated our models using a dataset of ultrasound images consisting of four types of thyroidal images: autoimmune, nodular, micro-nodular, and normal. Results: Excellent results were obtained by the ensemble CNN-VGG method, which outperformed the 5-CNN and VGG-19 models: 97.35% for the overall test accuracy with an overall specificity of 98.43%, sensitivity of 95.75%, positive and negative predictive value of 95.41%, and 98.05%. The micro average areas under each receiver operating characteristic curves was 0.96. The results were also validated by two physicians: an endocrinologist and a pediatrician. Conclusions: We proposed a new deep learning study for classifying ultrasound thyroidal images to assist physicians in the diagnosis process.

Highlights

Autoimmunity is related to the pathogenesis of many thyroid diseases, including hyperthyroidism Graves’ disease, hypothyroidism with autoimmune or Hashimoto’s thyroiditis, asymptomatic and postpartum thyroiditis, and some forms of neonatal thyroid dysfunction [1]
Approval was obtained from the institutional review boards of all institutions, and requirement for informed consent was obtained as the study design was based on a prospective research of medical tests and ultrasound images
The first method we proposed, called 5-convolution neural networks (CNNs), was an efficient lightweight architecture based on CNN network and trained in an end-to-end manner on the train dataset of 2297 US

Summary

Introduction

Autoimmunity is related to the pathogenesis of many thyroid diseases, including hyperthyroidism Graves’ disease, hypothyroidism with autoimmune or Hashimoto’s thyroiditis, asymptomatic and postpartum thyroiditis, and some forms of neonatal thyroid dysfunction [1]. Autoimmune hypothyroidism (AH) is usually divided into goiter (Hashimoto’s thyroiditis (HT)) and non-thyroid primary edema. Thyroid disorders have a great incidence in the worldwide population, so the development of alternative methods for improving the diagnosis process is necessary. Materials and Methods: For this purpose, we developed an ensemble method that fused two deep learning models, one based on convolutional neural network and the other based on transfer learning. We trained and validated our models using a dataset of ultrasound images consisting of four types of thyroidal images: autoimmune, nodular, micro-nodular, and normal. Results: Excellent results were obtained by the ensemble CNN-VGG method, which outperformed the 5-CNN and VGG-19 models: 97.35% for the overall test accuracy with an overall specificity of 98.43%, sensitivity of 95.75%, positive and negative predictive value of. The results were validated by two physicians: an endocrinologist and a pediatrician

Objectives

Methods

Results

Discussion

Conclusion