Angle-dependent ultrasonic detection and imaging of two types of brachytherapy seeds using singular spectrum analysis

Abstract

Brachytherapy to treat prostate cancer uses transrectal ultrasound to guide implantation of titanium-shelled radioactive seeds. Transperitoneal implantation allows errors in placement that cause suboptimal dosimetry. Conventional ultrasound cannot reliably image implanted seeds; therefore, seed misplacements cannot be corrected in the operating room. Previously, an algorithm based on singular spectrum analysis was shown to image palladium seeds better than B-mode ultrasound could. The algorithm is now applied to imaging an iodine seed in gel and in beef tissue as a function of seed angle relative to the incident ultrasound. Results indicate that both seed types are imaged reliably by the algorithm.

Introduction

This letter discusses a candidate method of imaging brachytherapy seeds in real-time using transrectal ultrasound (TRUS). Brachytherapy is a procedure used to treat gland-confined prostate cancer by permanent implantation of radioactive seeds within the prostate gland; currently, the most-commonly used seeds contain radioactive iodine or palladium.1, 2 These seeds have a thin, rigid, cylindrical, ultrasonically specular, titanium shell that is 4.5 mm long by 0.8 mm in diameter and TRUS is the standard imaging modality for treatment-planning dosimetry, guiding, and monitoring seed implantation.3, 4 Seeds are inserted transperitoneally through a needle attached to the TRUS probe. After insertion, the seeds ideally would be oriented normally to the ultrasonic beam and parallel to the implanting needle axis. However, suboptimal seed insertion typically causes seeds to rotate slightly so they are not normal to the transducer beam. Because the seeds are small and highly specular, they cannot be reliably imaged with TRUS if they are not normal to the ultrasound beam axis.

In two previous publications, we described development and evaluation of a seed-imaging algorithm based on a signal-processing framework called singular spectrum analysis (SSA).5, 6 The algorithm uses only echo-signal envelopes to generate color-coded images termed P-mode images. The P-mode images are indicative of the likelihood of seed presence as quantified by the algorithm. Evaluations of P-mode images showed them to be quantitatively superior to conventional ultrasound B-mode images to detect seeds. In the previous studies, all experiments were conducted on a popular type of palladium-based seed. This letter describes the performance of the SSA algorithm for imaging a different, popular type of iodine-based seed. SSA algorithm performance was evaluated as the iodine seed orientation was varied with respect to the transducer. Ultrasound experiments were conducted with seeds embedded in an ideal, clutter-free acoustical environment (i.e., acoustically transparent gel pad) and in a tissuelike, clutter-containing environment (i.e., degassed ex vivo beef). Iodine-seed experiments were performed in a manner identical to the previously reported experiments conducted with the palladium seeds. We also performed iodine-seed simulations based on empirical data acquired from the ideal-environment experiments. Score metrics previously developed to evaluate palladium-seed imaging performance were used to quantify iodine-seed imaging performance.6

Material and methods

The two types of seeds used in this study were devoid of radioactive content. The first type of seed was a commonly used version of radioactive palladium (Pd¹⁰³) seeds.7, 5, 6 The second type of seed was a commonly used version of radioactive iodine (I¹²⁵) seeds. The palladium and iodine seeds are termed Pd-103 and I-125 seeds, respectively. New data were acquired with I-125 seeds and compared to data previously published and obtained with Pd-103 seeds.5, 6 Both types of seeds consist of cylindrical thin titanium shells having the standard outer dimensions of brachytherapy seeds: 4.5 mm in length and 0.8 mm in diameter. However, the I-125 seed has convex rounded ends8 and the Pd-103 seed has concave cupped ends. The I-125 seed has radioactive material coated on a 3-mm-long silver rod centered within its titanium shell and the Pd-103 seed has radioactive material coated on two graphite cylinders separated by a lead cylinder.5

Experimental data were collected using exactly the previously published protocol.6 Briefly, data were acquired with a 5-MHz single-element transducer (Panametrics Inc., Waltham, MA). The transducer had a focal length of 51 mm and a diameter of 14 mm and was excited with a Panametrics 5900 pulser∕receiver unit. The radio-frequency (rf) echo signals were digitized at a sampling rate of 50 MS∕s, and the spacing between adjacent A-lines was 100 μm. Gel-pad and ex vivo beef experiments were conducted using I-125 seeds in the same manner as the Pd-103 experiments were previously conducted,6 with scanning performed in transverse and longitudinal planes and with angles between the beam axis and the normal to the seed axis varying from 0 to 22 deg. (Longitudinal-scan data were acquired by scanning in a plane parallel to the long axis of the seed and transverse-scan data were acquired by scanning in a plane perpendicular to the plane in which the seed was tilted.)

The simulation methods previously used for Pd-103 seeds were used to simulate rf echo-signal data from an I-125 seed.6 I-125 echo data was simulated for tilt angles ranging from 0 to 10 deg in both longitudinal and transverse seed orientations. The I-125 simulations were based on empirical data obtained from the gel-pad experiments and designed so that a complete B-mode image could be formed. Simulation parameters for I-125 seeds were different for Pd-103 results. These differences are presented and discussed in Sec. 3.

Two types of scoring metrics were developed to evaluate the performance of the SSA algorithm to detect and image Pd-103 seeds at various angles.6 The same metrics were used to quantitatively compare the ability of the SSA algorithm to detect I-125 and Pd-103 seeds. The first type of scoring metric, termed Score, was developed specifically for gel-pad experiments because no false-positives were present. The second type of scoring metric was developed to evaluate performance in beef experiments and to account for possible false-positives as well as false-negatives. The two measures second type were termed seed score (SS) and worst false positive (WFP); SS quantified how well the seed was detected and imaged while WFP quantified possible false-positives. The scores (i.e., WFP and SS) were computed from the P-mode images formed by the SSA algorithm and also from the conventional B-mode images. The subscripts P or B indicated scores based on P-mode and B-mode images, respectively. Computing scores for P-mode and B-mode images allowed quantitatively comparing the performance of the SSA algorithm with respect to the performance of conventional B-mode images.

Results

The longitudinal and transverse scans of a single I-125 seed in a gel pad were analyzed with the SSA algorithm. The simulations of rf echo signals from I-125 seeds were derived from experimental data and then used to generate P-mode images. To evaluate simulation performance, each angle (i.e., 0 to 10 deg) in each direction was simulated ten times with different realizations of white Gaussian noise. The white Gaussian noise amplitude was chosen to match the experimental signal-to-noise ratio (SNR) for the same angle and orientation. Experimentally, the SNR varied from 46 to 27 dB. The largest SNR was found for the longitudinal orientation at 0 deg and the smallest SNR for the transverse orientation at 10 deg.

Figure 1 presents the results obtained from gel-pad experiments. All the P-mode images (Figs. 1a1–a4 and 1b1–b4) are displayed with the same color scale to facilitate visual comparison. These P-mode images indicate that for both orientations and both angles, the I-125 and Pd-103 are perfectly detected and imaged by the SSA algorithm. The most striking difference between the P-mode images of I-125 and Pd-103 occur in the actual P-values. The P-values always are higher for the I-125 seed as shown by the dark red compared to the orange present on the P-mode images at 0 deg (i.e., Fig. 1a1 compared Fig. 1b1 and Fig. 1a3 compared Fig. 1b3) and the orange compared to the blue on the P-mode images at 4 deg (i.e,. Fig. 1a2 compared Fig. 1b2 and Fig. 1a4 compared Fig. 1b4). (The tilt angle of 4 deg was chosen to illustrate the differences between the two types of seeds because for this modest angle, both seed types should be reliably imaged. Therefore, performance differences should be apparent at this angle.)

Figure 1.

(a1)–(a4): P-mode images of Pd-103 seed in a gel pad at transverse and longitudinal orientations of 0 and 4 deg. (b1)–(b4): P-mode images of I-125 seed in gel pad at transverse and longitudinal orientations of 0 and 4 deg. (c) and (d) rf and envelope of central A-lines. (e) P-line obtained by processing the central A-lines with the SSA algorithm.

To investigate why the I-125 seed yields larger P-values (i.e., is better detected by the SSA algorithm), the central A-lines of Figs. 1a1 and 1b1 are displayed in Figs. 1c (time-delayed for visualization). Both A-lines contain a main signal followed by a repetition signal about 0.8 μs later. The main-signal amplitudes are nearly identical for both seeds (<3.3% relative difference), but the repetition-signal amplitude of the I-125 seed is approximately 44% greater than that of the Pd-103 seed. This difference is clear in the envelope of the two A-lines shown in Fig. 1d. The envelope of the repetition signal from the I-125 seed is 5 dB greater than that from the Pd-103 seed. Figure 1e displays the resulting P-line obtained by processing the central A-lines of Fig. 1c with the SSA algorithm. It reveals that the peak P-value for the I-125 seed is 12 dB greater than that for the Pd-103 seed. The SSA algorithm uses envelope signals to quantify the signal repetition.5 Therefore, the SSA algorithm is expected to yield greater P-values for the I-125 seeds than for the Pd-103 seeds because of the greater repetition-signal amplitude of the I-125 seed. This greater repetition-signal amplitude was incorporated in the I-125 seed simulations. These simulations were performed in the same manner as for the Pd-103 seed as described in detail previously.6 The empirical values from the I-125 seed gel-pad experiments were used to determine the amplitudes of the pulses used to simulate I-125 seed RF echo signals.

Figures 2a, 2b display the scores obtained for I-125 and Pd-103 for simulations and gel-pad experiments. For the simulations, each data point was the average of 10 realizations, displayed with error bars representing the standard deviations. From 0 to 4 degrees, the experimental longitudinal scores for the I-125 seed were between 8 and 12 dB greater than those for the Pd-103 seed. At angles greater than 5 degrees, the I-125 and Pd-103 experimental scores become nearly identical and reached a plateau near 85 dB. The simulated scores for I-125 seeds in the longitudinal orientation had similar values to the corresponding experimental scores except for the low value at 0 degrees. At angles smaller than 5 degrees, the simulated Pd-103 scores for the longitudinal orientation were about 8 dB higher than the experimental value, but again the trends were equivalent and the scores were nearly equal at angles greater than 5 deg. The experimental scores for the I-125 seed in the transverse orientation were 10 dB higher than those for Pd-103 seed from 0 to 8 degrees. At angles larger than 8 deg, the experimental transverse scores reached a plateau at about 45 dB. The scores from the simulated data in the transverse orientation are approximately the same for both types of seeds and follow the trend of the experimental data. The simulated I-125 scores are about 5 dB lower than the experimental values, and the simulated Pd-125 scores are about 5 dB higher than the experimental values.

Figure 2.

SSA-algorithm performance on simulations and experimental data. Scores for I-125 and Pd-103 seeds in a gel pad (a) from longitudinal-orientation scans and simulations and (b) from transverse-orientation scans and simulations. Seed scores and worst false positive scores from P-mode and B-mode images of a Pd-103 seed in ex vivo beef for (c) longitudinal- and (d) transverse-orientation scans and of an I-125 seed in ex vivo beef for (e) longitudinal- and (f) transverse-orientation scans.

As expected, visual inspection of the beef P-mode images (Fig. 3) reveals that algorithm performance worsens as the angle increases. For both orientations, imaging of the I-125 seed is satisfactory only at angles <9 degrees. When the angle of the I-125 seed increases beyond 10 deg, more false-positive P-values are visible in the surrounding tissue, and the P-values at the actual seed location decrease. Visual comparison of the P-mode images from both seed types reveals the same trends, but the P-mode images alone do not clearly reveal whether the SSA algorithm detects I-125 seeds better or worse than it detects Pd-103 seeds.

Figure 3.

Ex vivo experiments: P-mode images (40-dB dynamic range) of an I-125 seed inserted into beef. (a) and (b) Longitudinal- and transverse-orientation scans of the seed at various angles, respectively.

To quantify detection performance, Figs. 2c, 2d, 2e, 2f display SS and WFP from experimental ex vivo beef results. At 0 deg, the longitudinal SS_P for both seed types is approximately 68 dB. As the angle increases, SS_P gradually decreases to 27 and 40 dB for the I-125 and Pd-103 seeds, respectively. Nevertheless, for both seed types and at every angle, SS_P is greater than SS_B. In other words, the contrast provided by the P-mode image is greater than that provided by the B-mode image, i.e., seeds can be visualized better on P-mode than B-mode. However, at angles greater than 9 degrees, the SS_P value for Pd-103 is higher than the SS_P value for I-125. This result means that the algorithm would detect the Pd-103 seed better than the I-125 seed at these angles. The values for WFP_P and WFP_B are nearly equal for both seed types with values between 25 and 30 dB for WFP_P and between 10 and 13 dB for WFP_B. For the I-125 seed and at angles smaller than 14 deg, SS_P is greater than WFP_P. This result means that at angles smaller than 14 deg, the P-mode image of the SSA algorithm would detect a seed before any false positives. However, beyond 14 deg, approximately equal values of WFP_P and SS_P imply that the algorithm could produce false-positive seed detections. By comparison, at every angle SS_B is greater than WFP_B which means that no false-positives would occur by visually inspecting the B-mode images.

The transverse I-125 scores for the transverse orientation are similar to the corresponding Pd-103 scores. As shown in Fig. 2f, SS_P decreases quickly from 80 dB at 0 deg and remains constant at approximately 30 dB for angles beyond 5 deg. Furthermore, beyond 5 deg, WFP_P approximately equals SS_P. Similar to the longitudinal-orientation scores, for every angle (except the odd value at 16 deg), SS_P exceeds SS_B. These results indicate that the P-mode image has higher contrast than the B-mode image. Finally, at low angles (i.e., angles <3 deg), the SS_P value for the I-125 seed is about 10 dB higher than the SS_P value for the Pd-103 seed. Therefore, for near optimal insertion (i.e., at seed-tilt angles smaller than 3 deg), the SSA algorithm detects I-125 seeds better than Pd-103 seeds.

Discussion and conclusions

In this letter, we discuss the extension of a previously developed algorithm to image and detect brachytherapy seeds. The algorithm successfully imaged a popular type of palladium seed as described in prior publications.5, 6 This letter describes how the algorithm was applied to imaging a popular type of iodine seeds. Overall, results indicate that the algorithm successfully imaged the iodine seeds. The algorithm reliably imaged iodine seeds in ex vivo beef specimens at tilt angles smaller than 13 and 5 deg for longitudinal and transverse orientations, respectively.

Gel-pad experiments revealed that the repetition-signal amplitude of the I-125 seeds was greater than that of the Pd-103 seeds. The SSA algorithm exploits that repetition signals to make images; therefore, the algorithm detected I-125 seeds better than Pd-103 seeds in the gel pad. Although we remain unable to identify the mechanism that creates the repetition signal, the fact that a similar repetition signal was observed with two different types of seeds having a similar geometry (at least at 5 MHz) suggests that the repetition mechanism might be associated with the cylindrical shape of the seed. Furthermore, because the repetition-signal amplitude is greater for the I-125 seeds than for the Pd-103 seeds, signal differences may be attributable to differences in internal structure or details of shell shape. An encouraging conclusion of this study is that because both brachytherapy seeds produced a repetition signal, we were able to image these two different types of seeds with the same algorithm. This indicates that the algorithm may be robust to the type of seeds that are used, and this robustness could be very valuable clinically where many types of seeds can be encountered, even though the overall geometry (i.e., the diameter and length) of all seeds is well standardized.

Another valuable question that the results may answer is what type of seeds should be used clinically if the highest-priority criterion for seed selection is provision of the best possible feedback about seed location. Gel-pad experiments indicate that I-125 would constitute a better choice based on this criterion because I-125 consistently produced larger P-values. However, the ex-vivo beef experiments did not yield a conclusive answer. SS_P values were found to be greater for the I-125 seeds in the more-clinically transverse direction, but in the longitudinal direction, SS_P values were approximately the same for both seed types. Also, WFP_P values led to the conclusion that the SSA algorithm might produce slightly more false-positives when imaging I-125 seeds ex vivo than when imaging Pd-103 seeds for identical SS_P. Some of these findings can be explained by the experimental conditions. Although we carefully tried to acquire data for both seed types under identical experimental conditions, some minor differences existed. For example, the beef tissue was purchased from the same supplier and was of the same general type (i.e., cut) of beef, but the specimens for the Pd-103 and I-125 seeds were not identical.

The next step of this study will be evaluation of the SSA algorithm using clinical data. Because the SSA algorithm uses only signal envelopes, its translation to the clinic should be straightforward. We plan to evaluate its performance on a significant number of patients and to present performance in terms of receiver operator curves (ROCs).