Detecting Small Anatomical Structures in 3D Knee MRI Segmentation by Fully Convolutional Networks

Carl Petter Skaar Kulseng,Mengtao Sun,Kjell-Inge Gjesdal,Ibrahim A. Hameed,Li Lu

doi:10.3390/app12010283

Carl Petter Skaar Kulseng, Mengtao Sun + Show 3 more

Open Access

https://doi.org/10.3390/app12010283

Copy DOI

Abstract

Accurately identifying the pixels of small organs or lesions from magnetic resonance imaging (MRI) has a critical impact on clinical diagnosis. U-net is the most well-known and commonly used neural network for image segmentation. However, the small anatomical structures in medical images cannot be well recognised by U-net. This paper explores the performance of the U-net architectures in knee MRI segmentation to find a relative structure that can obtain high accuracies for both small and large anatomical structures. To maximise the utilities of U-net architecture, we apply three types of components, residual blocks, squeeze-and-excitation (SE) blocks, and dense blocks, to construct four variants of U-net, namely U-net variants. Among these variants, our experiments show that SE blocks can improve the segmentation accuracies of small labels. We adopt DeepLabv3plus architecture for 3D medical image segmentation by equipping SE blocks based on this discovery. The experimental results show that U-net with SE block achieves higher accuracy in parts of small anatomical structures. In contrast, DeepLabv3plus with SE block performs better on the average dice coefficient of small and large labels.

Highlights

Academic Editors: Manuel ArmadaKnee osteoarthritis is the most common musculoskeletal disease in the world [1].Lesions commonly induce structural changes within the small articular cartilage [2]
We propose four types of 3D U-net variants aimed at small anatomical structure segmentation in magnetic resonance imaging (MRI) images
We found that SE block performs well in small anatomical detection; Based on the success of SE block in U-net variants, we apply the DeepLabv3plus with SE block and transfer the 2D DeepLabv3plus into a 3D version for anatomical segmentation of real MRI images; In experiments, we improve the results from the small anatomical structure segmentation on the knee MRI images provided by Sunnmøre MR-Klinikk

Summary

Introduction

Knee osteoarthritis is the most common musculoskeletal disease in the world [1]. Lesions commonly induce structural changes within the small articular cartilage [2]. MRI (magnetic resonance imaging) technologies provide the means to characterise structural alterations in different joint tissues affected by osteoarthritis. MRI for presumed musculoskeletal disease can have unexpected vascular findings or pathology in the imaged field. Osteoarthritis is categorised by the progressive degradation of joint tissues with various abnormalities [4] and has been a severe issue in recent years. Some small anatomical structures around the joint are hardly detected. The veins and ligaments can show critical early alarms in musculoskeletal lesions [5]. This study will explore the performance of small structure segmentation in knee MRI by using deep learning

Methods

Results

Discussion

Conclusion