Capsule networks for segmentation of small intravascular ultrasound image datasets

Abstract

Purpose Intravascular ultrasound (IVUS) imaging is crucial for planning and performing percutaneous coronary interventions. Automatic segmentation of lumen and vessel wall in IVUS images can thus help streamlining the clinical workflow. State-of-the-art results in image segmentation are achieved with data-driven methods like convolutional neural networks (CNNs). These need large amounts of training data to perform sufficiently well but medical image datasets are often rather small. A possibility to overcome this problem is exploiting alternative network architectures like capsule networks. Methods We systematically investigated different capsule network architecture variants and optimized the performance on IVUS image segmentation. We then compared our capsule network with corresponding CNNs under varying amounts of training images and network parameters. Results Contrary to previous works, our capsule network performs best when doubling the number of capsule types after each downsampling stage, analogous to typical increase rates of feature maps in CNNs. Maximum improvements compared to the baseline CNNs are 20.6% in terms of the Dice coefficient and 87.2% in terms of the average Hausdorff distance. Conclusion Capsule networks are promising candidates when it comes to segmentation of small IVUS image datasets. We therefore assume that this also holds for ultrasound images in general. A reasonable next step would be the investigation of capsule networks for few- or even single-shot learning tasks.

3D pyramidal densely connected network with cross-frame uncertainty guidance for intravascular ultrasound sequence segmentation

“…The PDN-AC is superior to state-of-the-art IVUS segmentation networks IVUS-Net (Yang et al 2018), DPU-Net (Yang et al 2019), MFAU-Net (Xia et al 2020), (Nishi et al 2021) and (Bargsten et al 2021), according to table 4 and figure 10. These networks exploit individual 2D frames, ignoring longitudinal relations in IVUS sequences.…”

“…Modified from the IVUS-Net, a dual path U-Net (DPU-Net) with specific IVUS image augmentation was proposed to learn from a small training set (Yang et al 2019). (Xia et al 2020), (Nishi et al 2021), and (Bargsten et al 2021) introduced powerful U-Net, DeepLabV3, and CapsNet into IVUS segmentation with slight architecture modifications, respectively. A model cascading three modified U-Nets was presented by Li et al (2021).…”

Deep Learning Paradigm and Its Bias for Coronary Artery Wall Segmentation in Intravascular Ultrasound Scans: A Closer Look

“…The number of studies that fell under the intersection of (RBA, RBM, RBS), (RBA, RBS), and (RBM, RBS) was two, three, and one, respectively, whereas no common studies were found under intersection of (RBA, RBM). On the other hand, the number of studies under moderate bias (out of 60 studies) for RBM, RBA, and RBS was 9 (15%) [43,69,74,78,80,83,95,96,98], 20 (33%) [43,49,57,61,64,69,73,74,78,80,82,83,[86][87][88]95,98,120,160,161], and 18 (30%) [44][45][46][47][50][51][52][53][54][55][59][60][61]71,72,75,76,96], respectively. The number of studies that fell under the intersection of (RBA, RBS), (RB, MRBS), and (RBA,...…”

Optimal Deep Learning Based Atherosclerotic Plaque Classification on Intravascular Ultrasound Images

“…Bargsten et al [11] scientifically examined distinct capsule network framework variants and enhanced the accuracy of the segmentation of IVUS images. Later, the capsule network is compared to convolutional neural network (CNN) under different quantities of network parameters and training images.…”