ProLesA-Net: A multi-channel 3D architecture for prostate MRI lesion segmentation with multi-scale channel and spatial attentions

• Published in: Patterns (New York, N.Y.) Vol. 5; no. 7; p. 100992
• Main Authors: Zaridis, Dimitrios I., Mylona, Eugenia, Tsiknakis, Nikos, Tachos, Nikolaos S., Matsopoulos, George K., Marias, Kostas, Tsiknakis, Manolis, Fotiadis, Dimitrios I.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 12.07.2024; Elsevier
• Subjects: cancer detection; deep learning; magnetic resonance imaging; medical imaging; multi-scale attention; prostate lesion segmentation; deep learning; magnetic resonance imaging; cancer detection; multi-scale attention; prostate lesion segmentation; medical imaging
• ISSN: 2666-3899; 2666-3899
• DOI: 10.1016/j.patter.2024.100992

Prostate cancer diagnosis and treatment relies on precise MRI lesion segmentation, a challenge notably for small (<15 mm) and intermediate (15–30 mm) lesions. Our study introduces ProLesA-Net, a multi-channel 3D deep-learning architecture with multi-scale squeeze and excitation and attention gate mechanisms. Tested against six models across two datasets, ProLesA-Net significantly outperformed in key metrics: Dice score increased by 2.2%, and Hausdorff distance and average surface distance improved by 0.5 mm, with recall and precision also undergoing enhancements. Specifically, for lesions under 15 mm, our model showed a notable increase in five key metrics. In summary, ProLesA-Net consistently ranked at the top, demonstrating enhanced performance and stability. This advancement addresses crucial challenges in prostate lesion segmentation, enhancing clinical decision making and expediting treatment processes. •MSSE and MSAG deep-learning attention mechanisms are introduced and assessed separately•ProLesA-Net model for prostate lesion segmentation is proposed•The proposed model is compared to six deep-learning models on two datasets•The comparisons focus on small (<15 mm) and intermediate (15–30 mm) lesions Clinical identification of prostate lesions relies on biparametric MRI analysis. However, several technical limitations prevent the precise segmentation of lesions in the small size range. Deep learning, a type of machine learning consisting of several interconnected layers that process information, for example, MRI images, to generate informative outputs, is a promising strategy in the development of precise treatment strategies to assist MRI analysis and diagnostics. This paper presents a deep-learning framework to segment and identify small- to intermediate-sized prostate lesions. Of course, integrating deep-learning-based tools into clinical practice does not come without challenges, but architectures such as the one introduced in this work demonstrate the potential as well as the limitations of these technologies. This study presents ProLesA-Net, a deep-learning algorithm tailored for the segmentation of prostate lesions in biparametric MR images, leveraging multi-scale squeeze and excitation and attention gate mechanisms. Evaluated against existing multi-channel 3D segmentation models, ProLesA-Net achieves increased performance in detecting both small and intermediate lesions, thus enhancing the segmentation accuracy and efficiency of clinical diagnoses and treatment planning. Its increased performance, evidenced by testing on diverse datasets, marks an advancement in the application of artificial intelligence for medical imaging.