Reinforcement learning-based anatomical maps for pancreas subregion and duct segmentation

Amiri, Sepideh (author), Vrtovec, Tomaž (author), Mustafaev, Tamerlan (author), Deufel, Christopher L. (author), Thomsen, Henrik S. (author), Hylleholt Sillesen, Martin (author), Gudmann Steuble Brandt, Erik (author), Andersen, Michael Brun (author), Müller, Christoph Felix (author), Ibragimov, Bulat (author)

AbstractBackground: The pancreas is a complex abdominal organ with many anatom-ical variations, and therefore automated pancreas segmentation from medicalimages is a challenging application.Purpose: In this paper, we present a framework for segmenting individualpancreatic subregions and the pancreatic duct from three-dimensional (3D)computed tomography (CT) images.Methods: A multiagent reinforcement learning (RL) network was used to detectlandmarks of the head,neck,body,and tail of the pancreas,and landmarks alongthe pancreatic duct in a selected target CT image. Using the landmark detectionresults, an atlas of pancreases was nonrigidly registered to the target image,resulting in anatomical probability maps for the pancreatic subregions and duct.The probability maps were augmented with multilabel 3D U-Net architecturesto obtain the final segmentation results.Results: To evaluate the performance of our proposed framework, we com-puted the Dice similarity coefficient (DSC) between the predicted and groundtruth manual segmentations on a database of 82 CT images with manuallysegmented pancreatic subregions and 37 CT images with manually segmentedpancreatic ducts. For the four pancreatic subregions, the mean DSC improvedfrom 0.38, 0.44, and 0.39 with standard 3D U-Net, Attention U-Net, and shiftedwindowing (Swin) U-Net architectures, to 0.51, 0.47, and 0.49, respectively, whenutilizing the proposed RL-based framework. For the pancreatic duct, the RL-based framework achieved a mean DSC of 0.70, significantly outperformingthe standard approaches and existing methods on different datasets.Conclusions: The resulting accuracy of the proposed RL-based segmentationframework demonstrates an improvement against segmentation with standardU-Net architectures.

• Language: English
• Keywords: medical image analysis, landmark detection, image segmentation, pancreas region, deep learning, reinforcement learning
• Work type: Article
• Typology: 1.01 - Original Scientific Article
• Organization: FE - Faculty of Electrical Engineering
• Publication status: Published
• Publication version: Version of Record
• Year: 2024
• Number of pages: Str. 7378-7392
• Numbering: Vol. 51, iss. 10
• PID: 20.500.12556/RUL-164044
• UDC: 004.93:61
• ISSN on article: 2473-4209
• DOI: 10.1002/mp.17300
• COBISS.SI-ID: 202626819
• Publication date in RUL: 16.10.2024

Record is a part of a journal

• Title: Medical physics
• Shortened title: Med. phys.
• Publisher: American Institute of Physics
• ISSN: 2473-4209
• COBISS.SI-ID: 2725755

Licences

• License: CC BY 4.0, Creative Commons Attribution 4.0 International
• Link: http://creativecommons.org/licenses/by/4.0/
• Description: This is the standard Creative Commons license that gives others maximum freedom to do what they want with the work as long as they credit the author.

Secondary language

• Language: Slovenian
• Keywords: analiza medicinskih slik, razpoznavanje oslonilnih točk, segmentacija slik, trebušna slinavka, globoko učenje, spodbujevano učenje