Performance of Hardware Accelerated Bone Segmentation with Noise Reduction on 4DCT Images

Software, hardware and results generated from testing the performance of a bone segmentation algorithm implemented in both hardware and software for segmenting bone in medical images. The bone segmentation algorithm used was one described by Haas et al. (doi: 10.1088/0031-9155/53/6/017). The algorithm was implemented in full in software. Details of this implementation can be obtained from the following DOI: https://doi.org/10.15129/1a667dbc-8202-443d-a52b-45b5f8b498d2. A simplified version of this algorithm for processing image volumes was developed for hardware implementation, which was implemented in both hardware and software. The results of applying the simplified algorithm can be obtained from the following DOI: https://doi.org/10.15129/e46d23a5-227b-4e2b-8a51-0f4bdd17d644. The simplified algorithm was extended to improve its segmentation performance on noisy image data. This was again implemented in hardware and software and it is the data associated with this work that is presented here. The software algorithms were executed on a 2.6GHz Intel Core i5-3230M CPU, while the hardware system was implemented on a Xilinx Zynq Z7020 device on an AvNet ZedBoard development board. Image data in the form of 4-dimensional computed tomography (4DCT) scans of a Modus Medical QUASAR Programmable Respiratory Motion Phantom (Modus Medical Devices Inc. London, ON) were obtained from Edinburgh Cancer Centre (n=8) and used as the input data. Each of the datasets contained fifteen 3D image volumes that were each segmented. The time taken to process each image volume and each slice of each image volume was recorded for each of the three implementations. See README file for further details.