Robustness analysis of denoising neural networks for bone scintigraphy

• Published in: Nuclear instruments & methods in physics research. Section A, Accelerators, spectrometers, detectors and associated equipment Accelerators, spectrometers, detectors and associated equipment, 2022-09, Vol.1039, p.167003, Article 167003
• Main Authors: Kovacs, Akos, Bukki, Tamas, Legradi, Gabor, Meszaros, Nora J., Kovacs, Gyula Z., Prajczer, Peter, Tamaga, Istvan, Seress, Zoltan, Kiszler, Gabor, Forgacs, Attila, Barna, Sandor, Garai, Ildiko, Horvath, Andras
• Format: Article
• Language: English
• Subjects: Bone scintigraphy; Convolutional neural network; Deep learning; Image denoising; Machine learning; Medical imaging; Poisson noise
• ISSN: 0168-9002; 1872-9576
• DOI: 10.1016/j.nima.2022.167003

This paper describes and compares two neural network (NN) based noise filters developed for planar bone scintigraphy. Images taken with a gamma camera typically have a low signal-to-noise ratio and are subject to significant Poisson noise. In our work, we have designed a neural network based noise filter that can be used with planar bone scintigraphy recordings at multiple noise levels, instead of developing a separate network for each noise level. The proposed denoising solution is a convolutional neural network (CNN) inspired by U-NET architecture. A total of 1215 pairs of anterior and posterior patient images were available for training and evaluation during the analysis. The noise-filtering network was trained using bone scintigraphy recordings with real statistics according to the standard protocol, without noise-free recordings. The resulting solution proved to be robust to the noise level of the images within the examined limits. During the evaluation, we compared the performance of the networks to Gaussian and median filters and to the Block-matching and 3D filtering (BM3D) filter. Our presented evaluation method in this article does not require noiseless images and we measured the performance and robustness of our solution on specialized validation sets. We showed that particularly high signal-to-noise ratios can be achieved using noise-filtering neural networks (NNs), which are more robust than the traditional methods and can help diagnosis, especially for images with high noise content. [Display omitted] •Two neural network based noise filters have been designed.•They can be used with planar bone scintigraphy recordings at multiple noise levels.•They were trained on acquisitions created by the standard protocol.•The training and the evaluation method presented in this article does not require noiseless images.•The performance and robustness was measured on specialized validation datasets.