SUMNet: Fully Convolutional Model For Fast Segmentation of Anatomical Structures in Ultrasound Volumes

Abstract

Ultrasound imaging is generally employed for real-time investigation of internal anatomy of the human body for disease identification. Delineation of the anatomical boundary of organs and pathological lesions is quite challenging due to the stochastic nature of speckle intensity in the images, which also introduces visual fatigue for the observer. This paper introduces a fully convolutional neural network based method to segment organ and pathologies in ultrasound volume by learning the spatial-relationship between closely related classes in the presence of stochastically varying speckle intensity. We propose a convolutional encoder-decoder like framework with (i) feature concatenation across matched layers in encoder and decoder and (ii) index passing based unpooling at the decoder for semantic segmentation of ultrasound volumes. We have experimentally evaluated the performance on publicly available datasets consisting of 10 intravascular ultrasound pullback acquired at 20 MHz and 16 freehand thyroid ultrasound volumes acquired 11 − 16 MHz. We have obtained a dice score of 0.93 ± 0.08 and 0.92 ± 0.06 respectively, following a 10-fold cross-validation experiment while processing frame of 256 × 384 pixel in 0.035s and a volume of 256 × 384 × 384 voxel in 13.44s.

Adversarially Trained Convolutional Neural Networks for Semantic Segmentation of Ischaemic Stroke Lesion using Multisequence Magnetic Resonance Imaging

“…2(f) where finer details of core and penumbra are evident in our segmentation approach. Segmentation CNN: The segmentation CNN (net seg (·)) used is an encoder-decoder like architecture [12] with the encoder having layer definitions similar to that of VGG11 [13]. Concatenation of features across matched layers in the encoder and decoder is present in this architecture along with the passing of max pooling indices for up-sampling in the decoder.…”

Capsule networks for segmentation of small intravascular ultrasound image datasets

“…We chose a U-Net-like baseline CNN due to two reasons. First, previous work reports state-of-the-art results using encoder-decoder CNNs [7,15,17,19,28,29,31]. Second, the capsule network also features an encoder-decoder structure which makes both networks more comparable.…”

“…The authors of [4], e.g., combine an ensemble support vector machine pixelwise classifier with a deformable model to extract lumen and media-adventitia borders. Approaches using CNNs mainly rely on encoder-decoder architectures like U-Net [20] and report state-of-the-art results for segmentation of lumen and vessel wall [7,15,17,19,28,29,31]. However, CNNs depend heavily on the size of the underlying dataset as well as the quality of the corresponding annotations.…”

Catalyzing Clinical Diagnostic Pipelines Through Volumetric Medical Image Segmentation Using Deep Neural Networks: Past, Present, & Future

“…In contrast, classic high-level methods take more information into account and are able to model the geometry and physics of the target anatomy. Unfortunately, these models have historically been laborious to implement as modeling of those parameters necessitates feature engineering which in turn relies on a priori knowledge about the speckle patterns of regions of interest as well as organ geometry [51]. In response, researchers have investigated using deep neural networks, especially in 3D space, to automatically extract features to drive classical methods in a hybrid framework.…”