Pediatric Pancreas Segmentation from MRI Scans with Deep Learning

Abstract: Objective: Our study aimed to evaluate and validate PanSegNet, a deep learning (DL) algorithm for pediatric pancreas segmentation on MRI in children with acute pancreatitis (AP), chronic pancreatitis (CP), and healthy controls. Methods: With IRB approval, we retrospectively collected 84 MRI scans (1.5T/3T Siemens Aera/Verio) from children aged 2-19 years at Gazi University (2015-2024). The dataset includes healthy children as well as patients diagnosed with AP or CP based on clinical criteria. Pediatric and general radiologists manually segmented the pancreas, then confirmed by a senior pediatric radiologist. PanSegNet-generated segmentations were assessed using Dice Similarity Coefficient (DSC) and 95th percentile Hausdorff distance (HD95). Cohen's kappa measured observer agreement. Results: Pancreas MRI T2W scans were obtained from 42 children with AP/CP (mean age: 11.73 +/- 3.9 years) and 42 healthy children (mean age: 11.19 +/- 4.88 years). PanSegNet achieved DSC scores of 88% (controls), 81% (AP), and 80% (CP), with HD95 values of 3.98 mm (controls), 9.85 mm (AP), and 15.67 mm (CP). Inter-observer kappa was 0.86 (controls), 0.82 (pancreatitis), and intra-observer agreement reached 0.88 and 0.81. Strong agreement was observed between automated and manual volumes (R^2 = 0.85 in controls, 0.77 in diseased), demonstrating clinical reliability. Conclusion: PanSegNet represents the first validated deep learning solution for pancreatic MRI segmentation, achieving expert-level performance across healthy and diseased states. This tool, algorithm, along with our annotated dataset, are freely available on GitHub and OSF, advancing accessible, radiation-free pediatric pancreatic imaging and fostering collaborative research in this underserved domain.

Deep Learning for Pediatric Pancreas Segmentation in MRI Imaging

The work titled "Pediatric Pancreas Segmentation from MRI Scans with Deep Learning" explores a noteworthy endeavor in medical imaging by applying deep learning (DL) methodologies to the challenge of pediatric pancreas segmentation. While segmentation techniques in adult CT and MRI have evolved considerably, pediatric applications present unique hurdles such as small organ size, significant morphological variability, and limited contrast between the pancreas and surrounding tissues on MRI scans. The authors of the study introduce PanSegNet, a DL-based tool tailored specifically for the segmentation of the pediatric pancreas across multiple disease states as well as healthy controls. Their choice of the programming environment, dataset details, and the outcomes of their tool constitute a substantive contribution to the domain.

Methodology and Dataset

Utilizing a retrospective collection of 84 MRI scans from children aged 2-19 with normal health, acute pancreatitis (AP), or chronic pancreatitis (CP), the study underscores the importance of dataset diversity and representation. The annotation quality was ensured through manual segmentations by pediatric and general radiologists corroborated by an experienced senior pediatric radiologist. PanSegNet's architecture incorporated a linear self-attention mechanism within a nnUNet framework, optimized to manage large MRI datasets and fine structural details. The dataset was curated considering essential imaging protocols like T2-weighted sequences, leveraging Siemens Aera/Verio 1.5T/3T scanners.

Results

PanSegNet affirmed its capacity with Dice Similarity Coefficient (DSC) scores of 88% in healthy controls, 81% in AP, and 80% in CP. These figures surpass current norms reported in such demographic studies, indicating reliable segmentation across varied pathological manifestations. Predictions aligned well with manual assessments, evidenced by R2 values of 0.85 for healthy subjects and 0.77 for those with disease. Despite challenges in capturing subtle morphological changes, primarily caused by anatomical variances and fluid-related artifacts present in CP and AP cases, the model demonstrated strong clinical applicability.

Comparative Analysis

PanSegNet was juxtaposed against several state-of-the-art models — TransUnet, nnUNet3D, nnUNet2D, and SynergyNet, achieving predominantly superior DSC performance and a balanced HD95 metric. This benchmarking asserts the potential of using a focused single-organ segmentation approach over multi-organ methodologies within the pediatric landscape.

Implications and Future Prospects

The practical implications extend into clinical diagnostics where the integration of PanSegNet could offer nuanced volumetric analysis crucial for monitoring disease progression in pediatric populations prone to pancreatic ailments. Despite encountering segmentation failures in select cases with complex presentations, which highlighted the necessity for broadening and refining training datasets, the model exemplifies a significant step toward establishing AI-driven radiology norms.

The forthcoming trajectory of this research encompasses addressing the constraints of the current single-center dataset by incorporating multi-institutional collaborations and varying demographic representation (including diverse racial and ethnic backgrounds), alongside optimizing MRI protocols. Longitudinal studies evaluating the clinical impact on the management of pediatric pancreatitis and associated disorders will be vital for confirming the utility and accuracy of PanSegNet predictions in real-world scenarios.

In conclusion, the study by Keles et al. positions PanSegNet as a viable deep learning framework for advancing the domain of pediatric pancreas segmentation, promoting a template for future imaging-guided diagnostic frameworks in pediatric healthcare. The availability of source code and datasets encourages external validation and continuing evolution in the field, ensuring PanSegNet can meet the dynamic demands of clinical application and research collaboratively.