TY - GEN
T1 - Pituitary Gland Segmentation from Pre-processed Brain MRI Slice with VGG-UNet
AU - Kadry, Seifedine
AU - Yassine, Sahar
AU - Lin, Hong
AU - Rajinikanth, Venkatesan
AU - Gül, Ömer Melih
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - The Artificial Intelligence (AI)-supported data analysis is widely adopted in various domains to achieve better result for a chosen task. In medical domain, the AI-supported image analysis is commonly adopted to automate the image examination task. This research aims to propose a Deep Learning (DL)-based segmentation tool to extract the Pituitary Gland (PG) from the sagittal-plane brain MRI slice. The various stages in the proposed scheme includes: (1) image and mask collection from the repository, (2) three-dimension (3D) image to 2D image conversion using ITK-Snap and resizing, (3) pre-processing the MRI slice using Kapur’s Entropy and Butterfly Algorithm (KE + BA)-based thresholding, (4) implementing the VGG-UNet and extracting the PG with better accuracy, and (5) computing the necessary image metrics by comparing segmented PG with mask. This work implements the segmentation operation on the unprocessed and pre-processed MRI slices and verifies the performance of the implemented scheme based on the achieved image metrics. The experimental outcome authenticates that the VGG-UNet helps to achieve better Jaccard (91.37 ± 0.14), Dice (96.83 ± 0.04), and Accuracy (97.08 ± 0.02) compared to the unprocessed brain MRI slices. This confirms that the proposed DL-tool works well for the chosen image database.
AB - The Artificial Intelligence (AI)-supported data analysis is widely adopted in various domains to achieve better result for a chosen task. In medical domain, the AI-supported image analysis is commonly adopted to automate the image examination task. This research aims to propose a Deep Learning (DL)-based segmentation tool to extract the Pituitary Gland (PG) from the sagittal-plane brain MRI slice. The various stages in the proposed scheme includes: (1) image and mask collection from the repository, (2) three-dimension (3D) image to 2D image conversion using ITK-Snap and resizing, (3) pre-processing the MRI slice using Kapur’s Entropy and Butterfly Algorithm (KE + BA)-based thresholding, (4) implementing the VGG-UNet and extracting the PG with better accuracy, and (5) computing the necessary image metrics by comparing segmented PG with mask. This work implements the segmentation operation on the unprocessed and pre-processed MRI slices and verifies the performance of the implemented scheme based on the achieved image metrics. The experimental outcome authenticates that the VGG-UNet helps to achieve better Jaccard (91.37 ± 0.14), Dice (96.83 ± 0.04), and Accuracy (97.08 ± 0.02) compared to the unprocessed brain MRI slices. This confirms that the proposed DL-tool works well for the chosen image database.
KW - Brain MRI
KW - Evaluation
KW - Healthcare
KW - Kapur’s thresholding
KW - Segmentation
UR - https://www.scopus.com/pages/publications/105025001962
U2 - 10.1007/978-3-031-92143-8_4
DO - 10.1007/978-3-031-92143-8_4
M3 - Conference contribution
AN - SCOPUS:105025001962
SN - 9783031921421
T3 - EAI/Springer Innovations in Communication and Computing
SP - 49
EP - 61
BT - 8th EAI International Conference on Robotic Sensor Networks - EAI ROSENET 2024
A2 - Töreyin, Behçet Ugur
A2 - Köse, Hatice
A2 - Aydin, Nizamettin
A2 - Melih Gül, Ömer
A2 - Nimer Kadry, Seifedine
PB - Springer Science and Business Media Deutschland GmbH
T2 - 8th EAI International Conference on Robotics and Networks, EAI ROSENET 2024
Y2 - 3 September 2024 through 5 September 2024
ER -