Integration of Local and Global Features for Anatomical Object Detection in Ultrasound

Abstract

The use of classifier-based object detection has found to be a promising approach in medical anatomy detection. In ultrasound images, the detection task is very challenging due to speckle, shadows and low contrast characteristic features. Typical detection algorithms that use purely intensitybased image features with an exhaustive scan of the image (sliding window approach) tend not to perform very well and incur a very high computational cost. The proposed approach in this paper achieves a significant improvement in detection rates while avoiding exhaustive scanning, thereby gaining a large increase in speed. Our approach uses the combination of local features from an intensity image and global features derived from a local phase-based image known as feature symmetry. The proposed approach has been applied to 2384 two-dimensional (2D) fetal ultrasound abdominal images for the detection of the stomach and the umbilical vein. The results presented show that it outperforms prior related work that uses only local or only global features.

Automatic Detection and Measurement of Structures in Fetal Head Ultrasound Volumes Using Sequential Estimation and Integrated Detection Network (IDN)

“…The state-of-the-art techniques for finding anatomical structures in 3-D ultrasound images focus on automatic [7], [8] and semi-automatic [9] segmentation and detection [10], [11]. The most promising techniques are based on machine learning [10]- [13], pixel-wise classification [7], and deformable models [13], [14].…”

“…Although structural dependency in the model can improve robustness, these approaches have not yet been applied to fast and accurate pose estimation in 3-D images. The segmentation algorithms [7], [11]- [14] could, in principle, compute the measurements from the final segmentations, but they require the entire structure to be segmented which is slow and not as robust.…”

“…The approaches focus on designing specialized features [11], [27] and use discriminative learning to build robust detectors [12], [28]. Rahmatullah et al [28] automatically find stomach visualization plane in 3-D ultrasound volumes by searching for stomach bubble and umbilical vein landmarks.…”

“…The algorithm relies on AdaBoost training algorithm and detects the plane in 6 s on average with model trained on 2384 images. This computation time is lowered to under 1 s by using phase-based global features in the first level of the detector [11]. The algorithm only identifies an axis-aligned plane in a 3-D scan and therefore does not find the plane orientation.…”

“…These techniques could, in principle, be used to provide measurements of the structures after they have been segmented. The most reliable algorithms are based on pixel-wise classifiers [7], machine learning [11], [12], and deformable models [13], [14]. Gooding et al [9] propose a semi-automatic method for segmenting ovarian follicles.…”