Deep learning based diagnosis for cysts and tumors of jaw with massive healthy samples

Dan Yu,Huiyong Zhu,Jiacong Hu,Mingli Song,Zunlei Feng

doi:10.1038/s41598-022-05913-5

Abstract

We aimed to develop an explainable and reliable method to diagnose cysts and tumors of the jaw with massive panoramic radiographs of healthy peoples based on deep learning, since collecting and labeling massive lesion samples are time-consuming, and existing deep learning-based methods lack explainability. Based on the collected 872 lesion samples and 10,000 healthy samples, a two-branch network was proposed for classifying the cysts and tumors of the jaw. The two-branch network is firstly pretrained on massive panoramic radiographs of healthy peoples, then is trained for classifying the sample categories and segmenting the lesion area. Totally, 200 healthy samples and 87 lesion samples were included in the testing stage. The average accuracy, precision, sensitivity, specificity, and F1 score of classification are 88.72%, 65.81%, 66.56%, 92.66%, and 66.14%, respectively. The average accuracy, precision, sensitivity, specificity, and F1 score of classification will reach 90.66%, 85.23%, 84.27%, 93.50%, and 84.74%, if only classifying the lesion samples and healthy samples. The proposed method showed encouraging performance in the diagnosis of cysts and tumors of the jaw. The classified categories and segmented lesion areas serve as the diagnostic basis for further diagnosis, which provides a reliable tool for diagnosing jaw tumors and cysts.

Highlights

Since 2005, the World Health Organization (WHO) has labeled odontogenic keratocyst (OKC) as keratocystic odontogenic tumors (KCOTs) and has classified OKCs as tumors according to their behavior
What’s more, cysts (DCs and periapical cysts (PC)) achieve higher sensitivity/recall scores than tumors (ABs and KCOTs), which means that tumors are more likely to be misclassified
An encoder is pretrained on those massive healthy panoramic radiographs with self-supervised learning

Summary

Introduction

Huge, which heavily reduces the robustness and performance of the above transfer learning-based methods[18,23]. Sufficient labeled samples can effectively improve the performance of deep learning-based methods. Collecting and labeling massive lesion samples are time-consuming and heavily relies on the professional doctor’s experience. On the contrary, collecting massive healthy panoramic radiographs is more accessible and does not require a professional doctor’s annotation. The aim of this study is to develop an explainable and reliable method to diagnose cysts and tumors of the jaw with massive panoramic radiographs of healthy people based on deep learning. We develop a two-branch framework for diagnosing cysts and tumors of the jaw, where the position consistency constraint between the segmentation results and the response maps of classification is adopted to improve the reliability and explainability of the predicted results. Experiments show that the proposed two-branch network can simultaneously predict the category and area of lesion samples, which can serve as the diagnostic reference for further diagnosis of doctors

Objectives

Methods

Results

Discussion

Conclusion