A region of interest based approach for texture analysis of medical images in space–frequency domain. (May 2019)

• Record Type: Journal Article
• Authors: Pradhan, Pyari Mohan; Cheng, Chun Hing; Mitchell, Joseph Ross
• Is Part Of: Biomedical signal processing and control. Volume 51(2019)
• Page Start: 222
• Page End: 234
• Publication Date: 2019-05
• Subjects: Region of interest -- Texture analysis -- Medical images -- S-transform -- TT-transform -- Time–frequency representation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.02.014
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: FTFT-2D-ROI: Faster computation of average local spectrum and spectral statistics. Statistics: Mean and SD of SFR magnitude in Cartesian and polar coordinate systems. Simulation study is performed with rectangular and non-rectangular ROIs. Three skipping strategies employed for FTFT-2D-ROI. Computational performance of FTFT-2D-ROI is found to be superior to that of ST-2D. Low computation time leads to better real-time image analysis during virtual biopsy. Sensitivity analysis is performed to study the impact of parameters on FTFT-2D-ROI. Abstract: The two-dimensional S-transform (ST-2D) is a space–frequency representation (SFR) useful in image processing especially for texture analysis. The high computation time and inefficiency in computing the local spectrum at each pixel in an image motivated the authors to develop the two-dimensional fast time–frequency transform (FTFT-2D). However, in practice, the SFRs at individual pixels are seldom used for analysis. In most medical images, the spectral distribution over a region of interest (ROI) is of higher significance than that at individual pixels. This paper proposes a two-dimensional ROI-based fast time–frequency transform (FTFT-2D-ROI) for computing the average local spectrum and spectral statistics for an ROI faster and more efficiently using three skipping strategies. The spectral statistics for an ROI have been computed using mean and standard deviation of SFR magnitude in both Cartesian and polar coordinate systems. Simulations were performed with rectangular and non-rectangular ROIs while employing the three skipping strategies of FTFT-2D-ROI for a brain MRI image. Moving from skipping strategy I toward III, the computation time of FTFT-2D-ROI decreases significantly but at a cost of some degradation in accuracy of the SFR. The computational performance of the proposed FTFT-2D-ROI is also studied for different sizes of images, and found to be superior than that of the ST-2D. Further, sensitivity analysis was performed to study the impact of different parameters on the performance of the FTFT-2D-ROI. The end user can choose the skipping strategy based on the trade-off between accuracy and computation time.