Robust Resolution-Enhanced Prostate Segmentation in Magnetic Resonance and Ultrasound Images through Convolutional Neural Networks

Oscar J Pellicer-Valero,Paula Pelechano Gómez,Juan Luis Casanova Ramón-Borja,María Barrios Benito,María José Rupérez,José D Martín-Guerrero,Isabel Martín García,Victor Gonzalez-Perez,José Rubio-Briones

doi:10.3390/app11020844

Abstract

Prostate segmentations are required for an ever-increasing number of medical applications, such as image-based lesion detection, fusion-guided biopsy and focal therapies. However, obtaining accurate segmentations is laborious, requires expertise and, even then, the inter-observer variability remains high. In this paper, a robust, accurate and generalizable model for Magnetic Resonance (MR) and three-dimensional (3D) Ultrasound (US) prostate image segmentation is proposed. It uses a densenet-resnet-based Convolutional Neural Network (CNN) combined with techniques such as deep supervision, checkpoint ensembling and Neural Resolution Enhancement. The MR prostate segmentation model was trained with five challenging and heterogeneous MR prostate datasets (and two US datasets), with segmentations from many different experts with varying segmentation criteria. The model achieves a consistently strong performance in all datasets independently (mean Dice Similarity Coefficient -DSC- above 0.91 for all datasets except for one), outperforming the inter-expert variability significantly in MR (mean DSC of 0.9099 vs. 0.8794). When evaluated on the publicly available Promise12 challenge dataset, it attains a similar performance to the best entries. In summary, the model has the potential of having a significant impact on current prostate procedures, undercutting, and even eliminating, the need of manual segmentations through improvements in terms of robustness, generalizability and output resolution.

Highlights

In the field of medical imaging, segmentations are extremely useful for a plethora of tasks, including image-based diagnosis, lesion detection, image fusion, surgical planning or computer-aided surgery
Promise12 is an ongoing prostate segmentation challenge, wherein 50 Magnetic Resonance (MR) prostate images are provided along with their segmentation masks, and 30 additional images are provided without segmentations as a test set
The model was trained with five very challenging and heterogeneous MR prostate datasets with Ground Truth (GT) originating from many different experts with varying segmentation criteria

Summary

Introduction

In the field of medical imaging, segmentations are extremely useful for a plethora of tasks, including image-based diagnosis, lesion detection, image fusion, surgical planning or computer-aided surgery. Lesion detection [1], and the low complication profile [2], respectively, they are still not fully accepted in clinical guidelines. Accurate prostate segmentations are still hard and laborious to obtain, since they have to be manually annotated by expert radiologists and, even the interand intra-observer variability may be significant due to factors such as the lack of clear boundaries between neighboring tissues or the huge size and texture variation of this gland among patients. Coefficient (DSC) and 1.5619 mm in terms of the Average Boundary Distance (ABD). Very similar results were obtained in [3], with experts achieving a DSC and an ABD of 0.83 and. Automatic segmentation algorithms for the prostate are increasingly sought-after

Methods

Results

Discussion

Conclusion