Manual Segmentation of Pituitary Gland and Surrounding Structures from T1-Weighed MR Images
Poster No:
2229
Submission Type:
Abstract Submission
Authors:
Mubaraq Yakubu1, Alexander Hammers1, Andrew King1, Jonathan Shapey1, Qifan Chen1
Institutions:
1King's College London, London, United Kingdom
First Author:
Co-Author(s):
Introduction:
The pituitary gland, often termed the master gland, plays a crucial role in controlling other endocrine glands. Diseases of the pituitary gland not only disrupt the activities of other endocrine glands but can also affect neighboring structures, including compression of the optic chiasm. Pituitary adenomas are recognized as the primary pathology, ranking as the third most prevalent intracranial tumor. Magnetic Resonance Imaging (MRI) serves as the gold standard diagnostic tool for pituitary disease. The recent increase in automatic segmentation of medical images relies on manually annotated images as ground truth data, yet there are no established guidelines for such data concerning the pituitary gland and its surrounding structures.
This work intends to; 1) develop a comprehensive and validated methodology for manually delineating the normal pituitary gland and its surrounding structures, 2) generate ground truth data through the established approach to facilitate the training of automatic segmentation models, and 3) Lay the groundwork for the segmentation of pituitary disorders by establishing a foundational step based on the developed methodology and ground truth data.
This work intends to; 1) develop a comprehensive and validated methodology for manually delineating the normal pituitary gland and its surrounding structures, 2) generate ground truth data through the established approach to facilitate the training of automatic segmentation models, and 3) Lay the groundwork for the segmentation of pituitary disorders by establishing a foundational step based on the developed methodology and ground truth data.
Methods:
Two independent annotators utilized T1-Weighted MR images from the Hammers atlas database and Lyon database to manually annotate the pituitary gland, pituitary stalk, and optic apparatus (optic nerve, optic chiasm, and optic tract). A novel illustrated segmentation protocol (which is available at https://acrobat.adobe.com/id/urn:aaid:sc:EU:70f80d26-0e5b-4cce-89b2-0641072290e5) was developed and used as a guide for manual delineation of the region of interest by the annotators (Figure 1). ITK-SNAP software was used as the segmentation tool. Both Inter-rater and Intra-rater reliability tests were conducted. Results for Jaccard Index and Dice similarity coefficient were considered crucial in the evaluation of the performance of the segmentation. Other metrics such as false negative (FN) and false positive (FP) predictions and Hausdorff distance were also obtained.
Results:
Table 1 presents the Inter-rater reliability test scores for the two annotators; each having segmented at least 15 images. The intra-rater reliability test scores for annotator 1, based on 10 images, exhibited Jaccard index and Dice similarity coefficients of 0.82 and 0.90 for the pituitary gland, 0.71 and 0.82 for the pituitary stalk, and 0.75 and 0.85 for the optic apparatus, respectively. Annotator 2, working with 5 images, demonstrated Jaccard index and Dice similarity coefficients of 0.80 and 0.89 for the pituitary gland, 0.85 and 0.92 for the pituitary stalk, and 0.74 and 0.85 for the optic apparatus, respectively
Conclusions:
The pituitary gland is relatively small when compared to other brain structures, and the pituitary stalk is notably smaller. This makes it difficult to obtain remarkable similarity scores between the annotators. Despite this, the final scores obtained fall within acceptable range. The novel illustrated replicable protocol has potential to aid in segmenting T1-weighted MR images from different publicly available and local database which can enrich the available ground truth data for pituitary glands and its surrounding structures.
Modeling and Analysis Methods:
Segmentation and Parcellation
Neuroinformatics and Data Sharing:
Brain Atlases 2
Databasing and Data Sharing 1
Keywords:
MRI
Other - Pituitary and Manual Segmentation
1|2Indicates the priority used for review
Provide references using author date format
2. Grober, Y., Grober, H., Wintermark, M., Jane, J. A., & Oldfield, E. H. (2018). Comparison of MRI techniques for detecting microadenomas in Cushing's disease. Journal of neurosurgery, 128(4), 1051-1057.
3. Hammers, A., Chen, C. H., Lemieux, L., Allom, R., Vossos, S., Free, S. L., ... & Koepp, M. J. (2007). Statistical neuroanatomy of the human inferior frontal gyrus and probabilistic atlas in a standard stereotaxic space. Human brain mapping, 28(1), 34-48.
4. Mérida, I., Jung, J., Bouvard, S., Le Bars, D., Lancelot, S., Lavenne, F., ... & Costes, N. (2021). CERMEP-IDB-MRXFDG: a database of 37 normal adult human brain [18F] FDG PET, T1 and FLAIR MRI, and CT images available for research. EJNMMI research, 11(1), 1-10.
5. Molitch, M. E. (2017). Diagnosis and treatment of pituitary adenomas: a review. Jama, 317(5), 516-524.
6. Wang, S., Li, C., Wang, R., Liu, Z., Wang, M., Tan, H., ... & Zheng, H. (2021). Annotation-efficient deep learning for automatic medical image segmentation. Nature communications, 12(1), 5915.
7. Yao, A., Balchandani, P., & Shrivastava, R. K. (2017). Metabolic in vivo visualization of pituitary adenomas: a systematic review of imaging modalities. World neurosurgery, 104, 489-498.
8. Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., & Gerig, G. (2006). User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. Neuroimage, 31(3), 1116-1128.