(A) Animal Tissue and Phantom Studies and Calibrations -
(1) MR Animal tissues – Measurements and Discussions - The key Figures 2 & 3 were obtained from the tissue scan. In Figure 2, the `step-shape’ feature of the original digital data of the ERF of the animal tissue MR scan was reduced to the `smooth’ feature of the LSF by the effective Running Filter; And the averaging effect of the mid-point of the FWHM also contributed to both the sensitivity and accuracy. The length of the Running Filter may be lengthened with smoothening effect to reduce noise, or shortened to improve the sensitivity of the processing.. In Figure 3, the separation between the dotted line profiles indicated the extent of the “blurring” caused by the `penumbral-spread’ effect. It was measured to be 2.4 pixels across on average.
Discussion – As indicated by the following phantom calibrations, the de-convolution technique will pinpoint and reduce the separation of the `blurred’ dotted edge-profiles to the solid line profile (in Figure 3) e.g. from 2.4 pixels to 1/10th of a pixel in accuracy at the 1% low-contrast edge (Table 1 (Figure 32)). The well-resolved enhanced 0.3% ultra low-contrast rod in Figure 4 illustrated the high sensitivity of the methodology.
(2) Phantom studies - Measurements and Discussion –
Distance, Area, and Volume Calibrations, High Contrast & Low Contrast Resolution, Processing Time, and Random Noise Effects were studied:
(i) Distance, Area and Volume Measurements:-
Distance – mean accuracy to 1/50th of a pixel in dr (i.e. dr in dr/r) at 1% low-contrast edge; and, Area/Volume – Area - mean accuracy to 1/10th of a pixel in dr (i.e. 2prdr/pr2 or d(area) = 2prdr) and Volume = Area x Slice-thickness at 1% low-contrast edge (Table 1 (Figure 32)).
Discussion – Sub-pixel accuracy i.e. 1/10th of a pixel at the 1% low-contrast edge was achieved.
(ii) High Contrast Resolution:- Mean Modulation Transfer Function (MTF) at the high contrast Teflon edge were improved from 3.38 lp/cm to 7.75, 8.94, 9.61 and 10.04 lp/cm at 50% normalized modulation level and from 5.83 lp/cm to 11.89, 13.90, 15.03, and 15.75 lp/cm at 10% normalized modulation level respectively after 3x, 5x, 7x and 9x magnification and enhancement (Table 2 (Figure 33)). The calculations were made by the method of Judy P.F. (Ref. 1).
Discussion – Improved high-contrast resolution was achieved after magnification and enhancement.
(iii) Low-Contrast Resolution:-
(a) Resolved 1% low contrast rod down to 3mm rod at 50mA, 4 secs., 120 kVp, 5 mm slice (Table 2 (Figure 33)); or
(b) Well-resolved 0.3% low contrast to 2mm rod at 300mA, 120kVp, 4 secs., 10mm slice (Figure 4).
Discussion – the above demonstrates the improved sensitivity for the low-contrast enhancement.
(iv) Processing Time: At 3x, 5x, 7x, 9x magnification of Region of Interest (ROI) (i.e. 107x107, 64x64, 46x46, & 36x36 matrix respectively) took 4.67, 0.98, 0.43, 0.29 seconds by a 600 MHz CPU and 1.17, 0.25, 0.11, and 0.07 seconds by a 2.4 GHz CPU respectively (Table 3 (Figure 34)).
Discussion – Sub-second processing is possible for smaller ROI.
(v) The random noise effect on the enhanced linear edge-profile at an arm was reduced as indicated by da1 and da0 in Y=(a2 +/- da2)X2+(a1 +/- da1)X+(a0 +/-da0) from 57.50% to 12.4%, and 1.47% to 0.38% respectively by a Dynamic Filter. Enhancement and dynamic-filtered are illustrated by Figures 5-10.
Discussion – Noise effects on enhanced edge-profiles are proven random in nature. But this long Dynamic Filter is not applied to patient scans due to the `constraint’ artifact being introduced at the high spatial frequency regions.
Merits of the Methodology – Measurements & Discussion - The average “blurring” was measured to be 2.4 pixels on average across the ERF of an image edge-profile `before’ any enhancement by the De-Convolution Technique (Figure 2). By using this technique, the `step’ feature of the original digital data of the ERF of the above MR animal-tissue scan was reduced to the `smoothened’ feature of the LSF by the effectiveness of the Running Filter (Figure 2). The averaging effect of the mid-point of the FWHM helped further to contribute to both the sensitivity and the accuracy of the methodology, which is down to a sub-pixel level e.g. the edge detection and enhancement reduce the `blurring’ or `penumbral spread’ from an average 2.4 pixels to the sub-pixel level, i.e. 1/50th of a pixel for distance and 1/10th of pixel for area/volume measurements at the 1% contrast edges, making the implementation of the enhancement of an image up to 9 times the magnification possible.
This may potentially allow more accurate diagnosis and accurate & absolute quantitative measurements. The improved visibility of the 2mm and 3 mm rods of the enhanced 0.3% low-contrast phantom scan in Figure 4 indicated the improved sensitivity by using this methodology. As shown by the following examples, the `windows-settings’ dependency caused by both the Partial Volume Effect and the `Blooming’ Effect are also overcome by using this technique.
(B) Patient Field-Trial Examples-
(1) CT Liver tumor – Measurements and Discussions –
A Region of Interest (ROI) was magnified 7x. Measurements within tumor enclosure were: Area = 88.56 +/-1.96mm2; Volume = Area*Slice-thickness = 885.61 +/-19.55 mm3; Intensity = 88 CTU; Standard Deviation = 14 CTU; Intensity in the surrounding tissues outside of tumor = 112 CTU; Standard Deviation = 16 CTU. Figures 11-14 showed the original image with ROI centered on the tumor region, 7x ROI `before’ enhancement without edges demonstrated, 7x ROI `after’ enhancement with edges demonstrated, and 7x ROI `after’ enhancement with outlines for measurements. The outline of the tumor was traced out for the statistical measurements. The surrounding tissues’ statistics were also made for comparison.
Discussion - Sensitive and accurate measurements of properties within the low-contrast liver tumor edge-profile could aid evaluation of oncological therapy response. Ref. (2);
(2) A MR Meniscal tear – Measurements and Discussion –
A region of Interest (ROI) was centered on the tear of the Posterior Horn of the Medial Meniscus. Figures 15-18 showed the original image with ROI centered on the tear region. The 5x ROI `before’ enhancement without edges demonstrated, 5x ROI `after’ enhancement with edges demonstrated, and 5xROI enhanced with outlines for analyses. Measurements within the tear enclosure were:- Length of Tear = 16.07 +/- 0.01mm; Maximum Width = 2.29 +/- 0.01mm; Minimum Width = 1.37 +/- 0.01mm; Area of Tear Enclosure = 20.68 +/- 0.57mm2; Volume = Area*Slice-thickness = 72.38 +/- 1.98 mm3.
Discussion - The accurate measurements of minute effects within the MR knee images may be extended to other effects such as tumor, hairline fractures or bone fragments for oncological treatment response, lesion monitoring in the lungs, microsurgery, and radiotherapy treatment planning uses;
(3) Digital Fluorograph (DF) of femoral artery - Measurements and Discussions –
Figures 19-23 showed the various stages of `before’ or `after’ enhancement. Absolute measurements in DF in patient: Diameter of wire = 2.29±.02 pixels (=> 0.89mm).
Normal patient vessel lumen diameter=13.29±.02 pixels=>5.17±.01mm.
Patient vessel within stenosis cleared of blockage= 5.29 ±.02 pixels=>2.06±.01mm.
Stenosed vessel lumen diameter=(13.29-5.29)±.04 pixels=>3.11±.02mm.
Percentage diameter stenosis of superficial femoral artery = 3.11/5.17 =>60.15±0.84%.
Discussion - Through the calibration of using the 035J guide wire and the enhanced sharp edge-profiles, absolute measurements in DF may be obtained for various surgical or treatment uses, including, for example, the insertion of intravascular stents and angioplasty balloons.
(3) CT Lung tumor - Measurements and Discussion –
(a) Removal of the Partial Volume Effect (PVE):-
ROI was centered onto lung tumor and magnified 7x. Edge profiles were detected and enhanced. Mean angles of inclination at an edge profile in the z axis from contiguous pairs of images of Slices 14 & 15, and Slices 15 & 16 were +35.76+/-10.76o and -16.59+/-8.15o (Figure 24). The central image Slice 15 had slopes of reverse phases on both sides resulting in the partial volume effect (PVE) occurring within both the neighboring adjacent slices.
In the PVE region of Slice 15 (Figures 25-27): Area (PVE) = 107.49+/-2.51mm2; Mean Width = 4.48+/-.01mm; Length = 26.35+/-.01mm; Mean Intensity (M.I.) = -102+/-102 CTU; visible at WW/WL 1074/-150 (Figure 26) but invisible at WW/WL 238/70 (Figure 27). Within solid tumor: M.I.= 48+/-16 CTU, visible at both the above settings. In normal lung: M.I.= -603+/-89 CTU, invisible at WW/WL 238/70 (Figure 27).
Discussion - Partial Volume Effect Overcome - As described in the Evaluation Section, the detection of a slice of the lung tumor edge intensity is affected by the Partial Volume Effect and is windows-settings dependent. The Edge profiles of edge-dots derived by the De-Convolution Technique is windows-settings independent, i.e. free of the visibility restrictions of WW/WL;
(b) The Elimination of the `Blooming Effect’ -
Using the same patient image, the outer chest region was used for the study of the `Blooming Effect’. Figure 28 was a 9x ROI `before’ enhancement at WW=1801, C=168, showing the edge-dot profile lying outside the tissue region, whereas Figure 29 showing the edge-dot profiles lying within the tissue region at WW=1801, C=-530. But the `after’ enhancement images of Figures 30 & 31 showed the edge-dot profiles being on top of the true tissue edges irrespective of whether WW/C were set at 1801,168; or 1801,-530. Enhanced edge-profiles independent of the windows settings caused by the `Blooming Effect’ were obtained.
Discussion - The Elimination of the Blooming Effect - The enhanced edge-profile had the merit of removing the “Blooming Effect”, hence removing the windows-settings dependence feature.