Article
Application of a Convolutional Neural Network to kidney volume estimation
Search Medline for
Authors
Published: | February 26, 2021 |
---|
Outline
Text
Background: Artificial Intelligence (AI) is quickly advancing and economizing many medical procedures. It is particularly powerful when applied to complex but standardized data, such as radiology images [1]. We are applying AI to abdominal CT scans in order to segment the kidney and calculate its volume. Kidney volume is an important parameter in the diagnosis and therapy of renal diseases. The currently used, semi-automated kidney volume determination is time-consuming and prone to errors. A validated AI algorithm for kidney segmentation should significantly speed up the procedure and increase its accuracy.
Methods: A total of 2337 2D images were extracted at 1 cm distances from CT scans of 110 patient cases. The kidneys were manually segmented by radiology students and used to train a Convolutional Neural Network (CNN). The resulting kidney segmentation algorithm was used to predict kidney areas and calculate kidney volumes of 10 patient cases that were not used for training. Kidney volumes obtained in this way were compared to those acquired using the standard, semi-automated method on the same 10 patient cases using the Sectra imaging software [2].
Results: The preliminary CNN training and validation were performed on the right kidney. The training involved 50 rounds and took approximately 2 weeks. The method was specific, i.e. images that did not contain a kidney were confirmed as such. The entire sequence of segmentation and volume calculation with a trained CNN lasted, on average, 30 seconds per patient case. The mean ± SD kidney volumes were 151.2 ± 20.5 ml for the semi-automated standard method and 163.0 ± 32.4 ml for the CNN algorithm. The Pearson's correlation r value between volumes obtained with the methods was 0.87, with p (2-tailed) = 0.0012.
Conclusion: The AI method correctly identifies kidneys in CT images at a fraction of time required by the standard method. We are currently determining the method's accuracy using larger training and validation datasets and expanding it on the left kidney.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, Van Der Laak JA, Van Ginneken B, Sánchez CI. A survey on deep learning in medical image analysis. Medical image analysis. 2017;42:60-88.
- 2.
- Sectra Medical. Imaging IT Solutions That Lead The Way In Customer Satisfaction. [Accessed 2020 Mar 23].Available from: https://medical.sectra.com/