A quantitative analysis of two-dimensional manually segmented transrectal ultrasound axial images in planning high dose rate brachytherapy for prostate cancer

Background/Aim. Prostate delineation, pre-planning and catheter implantation procedures, in high-dose rate brachytherapy (HDR-BT), are commonly based on the prostate manually segmented transrectal ultrasound (TRUS) images. The aim of this study was to quantitatively analyze the consistency of prostate capsule delineation, done by a single therapist, prior to each HDR-BT fraction and the changes in the shape of the prostate capsule during HDR-BT, using two dimensional (2D) TRUS axial image. Methods. A group of 16 patients were treated at the Medical System Belgrade Brachytherapy Department with definitive HDR-BT. The total applied median dose of 52 Gy was divided into four individual fractions, each fraction being delivered 2–3 weeks apart. Real time prostate axial visualization and the manual segmentation prior to each fraction were performed using B-K Medical ultrasound. Quantitative analyses, analysis of an area and shape were applied on 2D-TRUS axial images of the prostate. Area analyses were used to calculate the average value of the cross-sectional area of the prostate image. The parameters of the prostate shape, the fractal dimension and the circularity ratio of the prostate capsule contour were estimated at the maximum axial cross section of the prostate image. Results. The sample group consisted of four phases, each phase being performed prior to the first, second, third and fourth HDRBT fraction, respectively. Statistical analysis showed that during HDR-BT fractions there were no significant differences in the average value of area, as well as in the maximum shape of prostate capsule. Conclusions. Quantitative analysis of TRUS axial prostate segmented images shows a successful capsule delineation in the series of manually segmented TRUS images, and the prostate maximum shape remaining unchanged during HDR-BT fractions.


Introduction
Prostate carcinoma is a common cancer and the second most common cause of cancer deaths in males 1 .Therapeutic management of prostate carcinoma has become complex, multidisciplinary and stage-specific 2,3 .Treatment options for patients with localized, organ confined prostate carcinoma range from observation, radical prostatectomy, radiation therapy (external and/or brachytherapy), hormonal therapy and various combinations of these [3][4][5] .
In practice, the final prostate contours for BT pre-planning or BT post-implantation planning could be obtained using manual, semi-automatic or automatic segmentation of the prostate margin images obtained by computed tomography (CT), magnetic resonance image (MRI) or transrectal ultrasound (TRUS) [18][19][20][21][22][23] .However, the prostate delineation, BT preplanning and catheter implantation BT procedures are commonly based on the use of manually segmented TRUS images obtained in axial and sagittal plains 6,15,22 (Figure 1).
Precise information about the size and the shape of prostate plays a key role in BT implantation and planning processes.Manual TRUS prostate capsule segmentation (delineation) is known to be difficult and hardly reproducible due to interference of numerous unfavorable factors 24,25 .Despite these downsides, TRUS in the prostate diagnostics value is not inferior to CT or MRI 24 .
An important issue in prostate BT is the quality and consistency of prostate segmentation by TRUS.It is well-known that human expert performance (urologist and/or radiation oncologist/brachytherapist) in manual segmentation, measured as the prostate BT workload, can affect final therapy result 26 , especially when fractionated high dose rate BT (HDR-BT) treatment planning is based on TRUS segmented prostate images.
However, when prostate HDR-BT is concerned, more than one implantation procedure could be applied.In these cases, it would be beneficial to have data on quantitative analysis of precision and consistency in prostate two dimensional (2D) TRUS axial image segmentation.Also, it would be beneficial to have precise information about the shape of the prostate during BT.Therefore the purpose of this study was to analyze consistency of the prostate capsule delineation prior to each HDR-BT fraction and the changes in the axial shape of the prostate capsule during HDR-BT using TRUS imaging.

Patients
A group of 16 patients, aged 65-82 years, in good health condition and performance status were treated at the Medical System Belgrade Brachytherapy Department, with definitive temporary HDR-BT (Microselectron HDR, ELECTA-Nucletron, the Netherlands).The total applied median dose was 52 Gy divided into four individual fractions, each fraction being delivered 2-3 weeks apart.The total HDR-BT dose was tailored individually, based on the selected association criteria of the American Brachytherapy Society 27 .The applied HDR-BT technique has been explained in detail elsewhere 4,28 .Androgen blockade therapy was applied to the group of patients 2-3 months prior to HDR-BT in order to intensify prostate encapsulation and shrink prostate volume (not necessarily uncirculated).

TRUS data acquisition
Patients were placed in lithotomy position under spinal anesthesia.Two parallel needles were positioned laterally to the prostatic urethra (Figure 1C) in order to prevent prostate movement during TRUS imaging and afterwards needle placement.Real time prostate axial visualization and manual segmentation prior each fraction was performed using B-K Medical Ultrasound (Model Hawk 2102) with a 7.5 MHz biplane TRUS probe mounted on a prostate-stepper.Manual segmentation, done by a single therapist, was performed from the prostate base (image b 0 ) to the prostate apex (b a ), using a 5•10 -3 m step (0 denotes a segment position from the prostate base and apex segment position).Prior to segmentation, Nucletron 5•10 -3 m grid was attached, centered and calibrated, allowing accurate rigid catheter/needle insertion.The median prostate volume calculated from TRUS data (V pc ) and volume (V p ) obtained by other means (CT or MRI), were 56•10 -6 m 3 (range 30-85•10 -6 m 3 ), and 43 cm 3 (range 30-68•10 -6 m 3 ), respectively, with the mean V pc /V p ratio of 1.4.The prostatic urethra volume was included in the calculated prostate volume.
For each patient, four imaging phases were completed and recorded: phase I -segmentation data collected before the first HDR-BT fraction; phase II -segmentation data collected before the second HDR-BT fraction; phase III -seg-mentation data collected before the third HDR-BT fraction; phase IV -segmentation data collected before the fourth HDR-BT fraction.A representative TRUS images prostate segment at the maximum axial cross sections (Figure 2) was used for HDR-BT pre-planning and catheter/needle application.The difference of maximum axial cross sections shown in Figure 2 can be directly related to the prostate volume changes between HDR-BT fractions.Post-therapy TRUS in the treated patients was not scheduled since this type of examination was not mandatory in HDR post-treatment evaluation protocol.

Image processing
All axial 2D-TRUS images were imported into the Image J, specialized public domain software for image analysis, developed by the National Institute of Health (USA, www.rsb.info.nih.gov/ij).Schematic representation of TRUS image processing procedure is illustrated in Figure 3.In the first step of the procedure grey-scale (Figure 3B) images were converted into the binary images (Figure 3A) using 'Threshold' tools.Figure 3B shows the silhouette area of the prostate capsule.In the second step of the procedure, binary images (Figure 3B) were converted into outline images (Figure 3B) using 'Outline' tools.Figure 3C shows the contour of the prostate capsule.

Area analysis of the TRUS axial image of the prostate capsule
The cross sectional area of the prostate capsule (A) was measured on a particular 2D-TRUS axial image.The area measurement was carried out using TRUS standard calculation algorithm by B-K Medical Ultrasound (Model Hawk 2102).The average value of the cross-sectional area of the prostate (A av ) for segmentation data collected before the first, second, third and the fourth HDR-BT fraction (phase I; phase II; phase III; phase IV) was calculated (Figure 4).

Shape analysis of the TRUS axial image of the prostate capsule
Shape analysis of the prostate was estimated by two parameters: the fractal dimension and the circularity ratio of the prostate capsule contour.The parameters of the prostate shape were estimated at the maximum cross section.These parameters were obtained by using Image J.

Fractal dimension of the prostate capsule contour
The fractal dimension of the prostate contour at the maximum axial cross-section (D max ) defines the irregularity in the shape of the image, showing shape deviation from the corresponding circle (D circle = 1.254) (Figure 5A).The image of the contour was analyzed by fractal analysis using the box-counting method 29,30 .This method consists of "covering" the image outline with sets of squares (Figure 5B).Each set is characterized by the size r of the square edge.The corresponding number of squares N that is necessary to cover the image outline is presented as a function of r (Figure 5C).D max is determined from the absolute slope value of the loglog relationship between N(r) and r (Figure 5C).In performing box-counting method, the box sizes were scaled to the base of 2; that is, 2 1 , 2 2 ... 2 k , where k continues until N is equal to unity 31 .Depending on the contour image size, the box-sizes were taken from 2 to 512 pixels.

Circularity of the contour prostate capsule
Although the prostate contours seem generally circular by nature, in practice we noticed a significant deviation of the prostate contours from the circle.A shape of the prostate capsule contour at the maximum axial cross section was estimated by using the circularity ratio (C max ).This parameter represents a measure of how particular shape deviates from a corresponding circle.If it is a circle contour, the circularity ratio is equal to one.In case of any other contour shape, the circularity ratio is less than one.This parameter is expressed as C = 4πAP -2 , where A is the area and P is the perimeter of the contour (Figure 6) 31 .

Statistical analysis
Statistical analysis of the calculated prostate parameters depends on whether the distribution is normal or not.In the  case of calculated parameters showing a normal distribution, each group is described by the corresponding mean value and standard error 32 .If the number of patients is smaller than 30, the character of distribution has to be tested by calculation of two statistical parameters: skewness, a 3 and excess of distribution, e 33 .Briefly, the intervals of skewness and excess values, which define the normal distribution, are estimated when these two parameters are divided by the corresponding mean square errors ( 3 and  4 ).If the absolute value of the quotients  3 /a 3 and  4 /e is less than or equal to 2, the data distribution can be considered as normal.Otherwise, the distribution of a sample (or population) has no characteristics of a normal distribution 34 .A statistical significance between mean values of parameters of prostate for 4-phase brachytherapy was estimated by Student´s t-test 34 .

Results
The sample group consisted of four phases, each phase being performed prior to the first (phase I), second (phase II), third (phase III) and the fourth (phase IV) BT fraction, respectively.Their representative axial images are shown in Figure 2. Table 1 shows value of the absolute ratios  3 /a 3 and  4 /e, for the average value of area (A av ), the fractal dimension (D max ) and the circularity ratio (C max ) for all four phases,.The absolute ratios ( 3 /a 3 ) and ( 4 /e) for the two parameters in all four of the phases are smaller than the critical value of 2. Therefore, the calculated values of three parameters showed normal distribution.Consequently, the values of three parameters of prostate capsule (A av , D max and C max ) can be expressed by the mean value and standards error.
The mean and standard errors of the A av , D max and C max for four phases were presented in Table 2.It was noticed that all phases had similar mean values of the three parameters of the prostate capsule.The statistical test estimated the difference between the parameters of mean values for 4-phase brachytherapy.
In the next step, the mean values of all possible pairs were tested by t-test (Table 3).The t-values of all three parameters are less than the t-values tabulated at the level of   1 significance p < 0.05 and p < 0.01 (Table 3).The t-test showed that there were no statistical differences between all possible pairs of the means A av , D max and C max , respectively (Table 3).Table 3 shows no significant difference in the values of A av , D max and C max , for phase I, phase II, phase III and phase IV, respectively.

Discussion
TRUS is used to guide placement of implants during prostate BT 21 , and as an input data for HDR-BT planning.This study quantitatively analyzed 2D-TRUS axial image obtained during prostate HDR-BT fractions.The average value area of cross section (A av ) and the two parameters relative to the maximum prostate capsule shape (D max and C max ) were determined by analyzing 2D-TRUS axial image.Statistical analysis showed no significant differences in the average value of cross sectionsal area during, HDR-BT fractions (phase I, phase II, phase III and phase IV) as well as in the maximum shape of prostate capsule (D max and C max ) (Table 3).
The absence of a significant difference in the average value of area (A av ) (Table 3) during prostate capsule delineation in the series of manually segmented TRUS images (phase I, phase II, phase III and phase IV) points to the consistency in prostate capsule delineation.This is in agreement with Solhjem et al. 19 , who demonstrated a significant reproducibility in prostate volume measurement using TRUS.However, manual TRUS prostate capsule delineation is known to be hardly reproducible due to the interference of numerous unfavorable factors 24,25 .This is the first time that the consistency in capsule delineation using TRUS was quantitatively proved, thus confirming a reproducibility of the manual TRUS segmented prostate axial images in preplanning and catheter/needle application for the HDR-BT.
On the other hand, our results show that the maximum prostate shape (D max and C max ) remains unchanged during analysis of HDR-BT fractions (Table 3).Holmes et al. 21onducted shape analysis of the prostate, however, the fractal dimension and the circularity were used for the first time in our study to quantify the shape of the prostate capsule.The fractal dimension is a geometrical feature of irregularly shaped objects 35 .Fractal analysis was applied to different medical images 36 , but there is no data on its application to the 2D-TRUS axial images of the prostate.Also, it is analyzed by means of the circularity, now commonly used as a parameter which quantifies the shape of two-dimensional objects 31 .
The lack of a significant difference in the shape of axial TRUS images during prostate capsule delineation in the series of manually segmented TRUS images (phase I, phase II, phase III and phase IV), could point to the absence of geometrical change in the fractions of HDR-BT.However, post HDR-BT data was not collected as TRUS in the treated patients.It was not mandatory in the HDR posttreatment evaluation protocol.Consequently, data for analysis of the prostate shape after the last treatment was not available.
The parameters of the cross-sectional prostate capsule (A av , D max and C max ) show that manually segmented TRUS images display good visibility necessary for precise delineation and monitoring of the prostate behavior during HDR-BT.
It was documented that prostate gland shrinkage may be expected in patients who undergo external beam radiotherapy shortly after starting androgen deprivation therapy 37 .Considering external beam radiotherapy alone, a particular effect of prostate gland shrinkage seems to be clinically insignificant 38 .

Conclusion
In this study, the applied quantitative analysis of transrectal ultrasound axial prostate segmented images obtained in each phase shows a successful capsule delineation in the series of manually segmented transrectal ultrasound images done by a single therapist and that the prostate maximum axial shape remains unchanged during the high-dose rate brachytherapy course, of approximately 2-month duration.This is the first time to determine the prostate capsule maximum shape (the fractal dimension and the circularity) of transrectal ultrasound axial images.These two prostate shape parameters may be applicable in quantification of prostate shape for the purpose of medical diagnostics and treatment.Transrectal ultrasound of the prostate is fast and relatively inexpensive diagnostic technique.Quantitative analysis showed that transrectal ultrasound of the prostate is accurate in pre-planning and catheter/needle application to high-dose rate brachytherapy.Additionally, the applied image analysis could be suitable for determination of the accuracy in other radiotherapy planning techniques.

Fig. 1 -
Fig. 1 -Typical transrectal ultrasound (TRUS) images of the prostate, shown with permission of the General Hospital Medical System Belgrade (MSB): A) sagittal plan, B) axial plan and C) segment axial plan at the maximum prostate cross-sectional area.

Fig. 2 -
Fig. 2 -Representative two dimensional transrectal ultrasound (2D-TRUS) images of the prostate at the maximum axial cross section for the individual patient: A) phase I; B) phase II; C) phase III; D) phase IV.

Fig. 3 -
Fig. 3 -Schematic representation of the two-dimensional transrectal ultrasound (2D-TRUS) axial image of the prostate capsule processing procedure: grayscale image (A), image of the silhouette area (B) and of contour (C).

Fig. 5 -Fig. 6 -
Fig. 5 -Illustration of calculating the fractal dimension of the prostate capsule contour, the fractal dimension of the circle is inscribed in the upper part of the figure (A).The application of the box-counting method to the contour, the whole image is covered with a set of squares and the squares which cover inner contour are counted (B).Log-log plot between the numbers of squares (N) and square size (r) is fitted by a straight line (C).The fractal dimension is 1.200, R 2 is the corresponding determination coefficient and p is the significance level.