From pixels to pathology: A novel dual-pathway multi-scale hierarchical upsampling network for MRI-based prostate zonal segmentation

Abstract

Prostate cancer is a prevalent and life-threatening disease characterized by abnormal cell growth within the prostate gland. Early and accurate diagnosis of prostate cancer is crucial for effective treatment planning. Magnetic Resonance Imaging (MRI) is valuable for diagnosing and assessing prostate cancer. Medical professionals use MRI to create segmentation for detecting prostate cancer. However, existing segmentation methods are limited in accurately delineating anatomical structures and tumor regions within the prostate. This research proposes an innovative methodology for advancing MRI-based prostate segmentation. The objective is to encompass anatomical and tumor zones within the prostate, facilitating precise diagnosis and treatment planning. The proposed dual-pathway multi-scale hierarchical upsampling network introduces significant modifications compared to the traditional UNet-based architecture. It outperforms previous studies, demonstrating superior performance in anatomical segmentation on both the ProstateX and Prostate158 datasets. It achieves an intersection over union of 0.8449 and a dice similarity coefficient of 0.9872 on ProstateX, as well as an intersection over union of 0.8065 and a dice similarity coefficient of 0.9831 on Prostate158, suppressing previous research by a significant margin. These results highlight the potential of this approach to advance the utility of MRI in diagnosing and planning the treatment of prostate-related pathologies, benefiting both patients and healthcare practitioners.

Keywords

• Prostate zonal segmentation
• Neural network
• Magnetic resonance imaging
• Deep learning