Identifying Periampullary Regions in MRI Images Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• The peri-ampullary (PA) region is critical for diagnosing and staging periampullary carcinoma (PAC).
• Accurate segmentation of the PA region in magnetic resonance imaging (MRI) is essential for effective clinical assessment.
• Manual segmentation by experts is time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a deep learning (DL) method for automated segmentation of the PA region in MRI images.
• To assess the performance of the DL algorithm in comparison to manual segmentation by expert radiologists.
• To evaluate the potential of DL for improving the efficiency and accuracy of PA region identification in preoperative MRI.

Main Methods:

• A dataset of MRI images from patients with and without PAC was curated.
• PA regions were manually annotated by expert radiologists.
• A DL model was trained, validated, and tested on randomly allocated patient data.
• Segmentation performance was quantified using the intersection over union (IoU) metric.

Main Results:

• The DL algorithm achieved high accuracy in segmenting the PA region on both T1 and T2 MRI sequences.
• The IoU scores for T1, T2, and combined T1/T2 images were 0.68, 0.68, and 0.64, respectively.
• The DL model demonstrated performance comparable to manual segmentation by human experts.

Conclusions:

• The developed DL algorithm shows significant promise for automated PA region segmentation in MRI.
• This non-invasive, automated approach can aid clinicians in precise identification and localization of the PA region.
• The DL method offers a potential improvement in preoperative MRI analysis for PAC patients.