Marker Myotendinous
3.6 Speckle Tracking Software Design and Implementation
3.6.4 Individual Error Analysis
One of the biggest problems found with all speckle tracking algorithms were the errors that arise due to the inconsistency of speckle patterns that comes and go between frames. Eventually, all algorithms fail to track the correct ROI along successive frames when more errors occur. Two types of error were identified:
• Stationary movement error
• Irregular movement error
The causes or these and the solutions adopted are discussed in the following sections.
3.6.4.1 Stationary Movement Error
The error happened when tracking showed no movement and the algorithm fails to find any match that met the threshold condition between the two frames. To solve this problem, the tracking algorithm proceeded with analysing the images from the next frame and marked if no movement was found and the previously successful matched image was used as the image reference. This procedure was repeated for the next five frames, and if no movement was found after five-frame duration, the software algorithm would terminate. If more errors were found during the tracking process the tracking would appear to be no movement at all as no possible matches were found throughout the whole image sequences. The study conducted was to understand better how the algorithm deals with unmatched criteria. The study also helped in identifying the optimum algorithm to be used in the later stage of the analysis and integrated into the tracking software system as the primary tracking algorithm. Understanding the frequency of the errors also contributed to the enhancement of the algorithm so that any errors in movement can be repaired by the use of previous best match as recovery mechanism during the tracking process.
3.6.4.2 Irregular Movement Error
An error happened when the algorithm fails to match between the two blocks correctly causing the tracking to track a different region throughout the whole tracking process. The movement of the tracked ROI can be seen as if it is jumping around erratically and sometimes the movement appeared to move further than it should move or move in the direction opposite the tracking path. The movement was considered to be an error when the tracked ROI jumped 15 pixels more than the previous tracked movement. This anomaly triggered the software algorithm to recheck the path of the tracking system and reanalysed the tracking process by selecting the next lowest (MSE) or highest (NCC) estimation values, and the procedure proceeds until the best location was found.
For counting the errors, every movement that exceeded 15 pixels from the previous tracking path was counted and averaged to get the mean value so that mean differences can be identified between algorithms and plotted to illustrate the difference. Pairwise comparisons were also used to the p value between the algorithms to differentiate further the interactions between the data collected for each algorithm. As the previous analysis, the errors found from this section did not represent the total errors for the algorithm in failing to track the tendon regions of highly speckled ultrasound images. The analysis conducted here contributes another half of the two principal errors identified, which influences the ability of the tracking algorithm to follow the movement.
Similar to the previous analysis, the idea of analysing the errors was to understand better how the algorithm deals with unmatched criteria. The analysis also helped in identifying the optimum algorithm to be used in the later stage of the analysis and integrated into the tracking software system as the primary tracking algorithm. Understanding the frequency of the errors also contributed to the enhancement of the algorithm so that any errors in movement can be recovered by the use of last best match as a recovery mechanism during the tracking process.
3.7 Summary
In summary, this chapter is a descriptive chapter where the methodology of data collection and analysis of the tracking algorithms are presented. The evaluation of the four algorithms and planned experiments for each study are described in the next chapter. The design and implementation of the speckle tracking software that was required to support these studies area was also described including the description of the graphical user interface (GUI) and the three tracking algorithms (NCC, MSE, LK) that were identified in the previous chapter. A new method is introduced, which is the combination of the algorithms NCC and MSE and therefore named as NCCMSE algorithm. With the use of SNR as the decision selector, these two algorithms are alternated to improve the tracking based on the value of the SNR. The performance of the four selected tracking algorithms will be described in the following Chapter.
The tracking accuracy as well as the computational cost of each algorithm was compared to the standard manual measurement. Finally, the reliability of the tracking algorithms are also discussed where two kinds of errors are being described as stationary movement errors and irregular movement errors. These reliability factors are evaluated in the next following chapter.