U-Net for Multi-Organ Segmentation of SPECT Projection Data
Mürschberger N, Reymann M, Cachovan M, Ritt P, Vija H, Maier A (2020)
Publication Language: English
Publication Type: Conference contribution, Abstract of a poster
Publication year: 2020
Conference Proceedings Title: U-Net for Multi-Organ Segmentation of SPECT Projection Data
Event location: Boston, Massachuchets
DOI: 10.1109/nss/mic42677.2020.9507779
Abstract
This work explores the usage of deep learning techniques for segmentation of anatomical structures on SPECT data. Previous works applied deep learning techniques on CT scans of anatomical information and afterwards registered the resulting segmentation mask on the SPECT data. We train a U-Net on 2D projection data extracted from Lu-177 MELP SPECT scans aim to predict the areas corresponding to liver, spleen and kidney. The challenging acquisition technique comes along with low resolution, low contrast between the abdominal organs and variations in size, shapes and location due to the differences in patients` anatomies. We use the dice similarity coefficient to evaluate the network. Visual validation, confusion matrix and recall shall evaluate the worthiness of further investigations in deep learning techniques used for nuclear medicine. We achieve a mean dice similarity coefficient of 72% and a validation loss of 11%, encouraging us to further investigate our approach.
Authors with CRIS profile
Maximilian Reymann
Lehrstuhl für Informatik 5 (Mustererkennung)
Philipp Ritt
Department of Nuclear Medicine
Andreas Maier
Lehrstuhl für Informatik 5 (Mustererkennung)
Involved external institutions
Germany (DE)How to cite
APA:
Mürschberger, N., Reymann, M., Cachovan, M., Ritt, P., Vija, H., & Maier, A. (2020, November). U-Net for Multi-Organ Segmentation of SPECT Projection Data. Poster presentation at IEEE, Boston, Massachuchets, US.
MLA:
Mürschberger, Nina, et al. "U-Net for Multi-Organ Segmentation of SPECT Projection Data." Presented at IEEE, Boston, Massachuchets Ed. IEEE, 2020.
BibTeX: Download