SpArNet: Sparse Asynchronous Neural Network execution for energy efficient inference
Convolutional neural networks for localization of partially inserted needles
Successful automated detection of the needle immediately after insertion is necessary to allow the physician identify and correct any misalignment of the needle and the targeat at early stages, which reduces needle passes and improves health outcomes. In our method, very short needles are detected with tip localization errors of less than 0.7 mm for lengths of only 5 mm and 10 mm at 0.2 mm and 0.36 mm voxel sizes, respectively.
A. Pourtaherian et. al, IJCARS 2018 IJCARS MICCAI best paper award
A. Pourtaherian et. al, MICCAI 2017
Mask-MCNet: instance segmentation in 3D point cloud data
A framework based on deep learning models for segmenting object instances in 3D point cloud data by predicting their 3D bounding box and simultaneously segmenting all the points inside each box.
F. Zanjani, A. Pourtaherian et. al, Neurocomputing 2020 (in press)
Gabor-based needle detection and tracking in three-dimensional ultrasound data volumes
A novel 3-D Gabor wavelet transformation is introduced and optimally designed for revealing the instrument voxels in the volume, while remaining generic to several medical instruments and transducer types. Experiments on diverse data sets, including in vivo data from patients, show that for a given transducer and an instrument type, high detection accuracies are achieved with position errors smaller than the instrument diameter in the 0.5-1.5-mm range on average.
A. Pourtaherian et. al, IEEE TMI 2017
A. Pourtaherian et. al, IEEE ICIP 2014
Fast planar segmentation of depth images
A fast planar object segmentation algorithm for depth images avoiding any surface normal calculation. This method first extracts 3D edges, then it localizes the lines falling between these edges and finally, it merges all the points on each pair of the intersecting lines into a plane.
H. Hemmat, A. Pourtaherian et. al, SPIE/IS&T EI 2015
TROD: tracking with occlusion handling and drift correction
A real-time tracking framework based on traditional computer vision and HOG features. An object detector is trained online in the first video frame and is used to localize the object in subsequent frames. Occlusion-handling and drift-correction techniques control the classifier update strategy over time to cope with object appearance changes.
A. Pourtaherian et. al, IEEE ICIP 2013