Annotated normal CT data of the abdomen for deep learning: Challenges and strategies for implementation

• Published in: Diagnostic and interventional imaging Vol. 101; no. 1; pp. 35 - 44
• Main Authors: Park, S., Chu, L.C., Fishman, E.K., Yuille, A.L., Vogelstein, B., Kinzler, K.W., Horton, K.M., Hruban, R.H., Zinreich, E.S., Fadaei Fouladi, D., Shayesteh, S., Graves, J., Kawamoto, S.
• Format: Journal Article
• Language: English
• Published: France Elsevier Masson SAS 01.01.2020
• Subjects: Abdominal computed tomography (CT); Artificial intelligence (AI); Image segmentation; Machine learning; Normal structures; Radiology; Image segmentation; Normal structures; Artificial intelligence (AI); Abdominal computed tomography (CT); Machine learning
• ISSN: 2211-5684; 2211-5684
• DOI: 10.1016/j.diii.2019.05.008

The purpose of this study was to report procedures developed to annotate abdominal computed tomography (CT) images from subjects without pancreatic disease that will be used as the input for deep convolutional neural networks (DNN) for development of deep learning algorithms for automatic recognition of a normal pancreas. Dual-phase contrast-enhanced volumetric CT acquired from 2005 to 2009 from potential kidney donors were retrospectively assessed. Four trained human annotators manually and sequentially annotated 22 structures in each datasets, then expert radiologists confirmed the annotation. For efficient annotation and data management, a commercial software package that supports three-dimensional segmentation was used. A total of 1150 dual-phase CT datasets from 575 subjects were annotated. There were 229 men and 346 women (mean age: 45±12years; range: 18–79years). The mean intra-observer intra-subject dual-phase CT volume difference of all annotated structures was 4.27mL (7.65%). The deep network prediction for multi-organ segmentation showed high fidelity with 89.4% and 1.29mm in terms of mean Dice similarity coefficients and mean surface distances, respectively. A reliable data collection/annotation process for abdominal structures was developed. This process can be used to generate large datasets appropriate for deep learning.