Name
Title | ||
|---|---|---|
Guidewire Segmentation in 4D Ultrasound Sequences Using Recurrent Fully Convolutional Networks. |
Abstract | ||
|---|---|---|
Accurate, real-time segmentation of thin, deformable, and moving objects in noisy medical ultrasound images remains a highly challenging task. This paper addresses the problem of segmenting guidewires and other thin, flexible devices from 4D ultrasound image sequences acquired during minimally-invasive surgical interventions. We propose a deep learning method based on a recurrent fully convolutional network architecture whose design captures temporal information from dense 4D (3D+time) image sequences. The network uses convolutional gated recurrent units interposed between the halves of a VNet-like model such that the skip-connections embedded in the encoder-decoder are preserved. Testing on realistic phantom tissues, ex vivo and human cadaver specimens, and live animal models of peripheral vascular and cardiovascular disease, we show that temporal encoding improves segmentation accuracy compared to standard single-frame model predictions in a way that is not simply associated to an increase in model size. Additionally, we demonstrate that our approach may be combined with traditional techniques such as active splines to further enhance stability over time. |
Year | DOI | Venue |
|---|---|---|
2020 | 10.1007/978-3-030-60334-2_6 | ASMUS/PIPPI@MICCAI |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 4 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Brian C. Lee | 1 | 0 | 1.69 |
| Kunal Vaidya | 2 | 0 | 0.34 |
| Ameet K. Jain | 3 | 0 | 0.34 |
| Alvin Chen | 4 | 28 | 4.23 |