DETECTING IMPLANTABLE MARKERS JUST 1.0MM IN DIAMETER

UMC Utrecht demonstrates automatic marker detection with amorphous-silicon imager

a-Si portal image showing lateralprostate with markers

Superior contrast resolution provided by an amorphous-silicon (a-Si) imager has enabled University Medical Center Utrecht investigators to detect implantable radiotherapy markers just 1.0 mm in diameter.

In a recent study,¹ the a-Si imager and related software facilitated rapid, accurate marker detection, and resulted in high marker detection success rates and localization accuracy. The center's imager technology and detection algorithms promise improved on-line target verification.

Enhanced beam delivery, through such methods as IMRT and advanced dose planning, has improved radiotherapy precision. But these sophisticated dose delivery mechanisms are ineffective if the dose is poorly targeted. While using fiducial markers has been an option to better target dose, their routine use has been impractical because current imaging technology requires an unacceptably large marker size of at least 1.6 mm.

Researchers at the University Medical Center Utrecht (UMCU) hypothesized that an a-Si imager could be harnessed to detect gold markers of a more clinically acceptable size.

'In radiotherapy, marker implantation has been used mostly to study the extent of tumor movement,' says Aart Nederveen, one of the UMCU study's principal researchers. 'For conformal therapy, marker-guided position verification is attractive because you can avoid errors due to set-up and organ motion (click for details). The problem has been that marker diameters couldn't be less than 1.6 mm and their implantation is very invasive and clinically unacceptable on a routine basis.'

Commercially available electronic portal imaging devices (EPIDs) are sophisticated enough to satisfy all requirements for on-line position verification with implanted fiducial markers, except for the clinically unacceptable large marker size required. For this reason, Mr. Nederveen selected an a-Si imager that could provide the needed contrast resolution to image markers smaller than 1.6 mm in diameter.

'We specified five requirements for our technique, the most important of which was marker size, which couldn't exceed 1.2 mm in diameter,' he notes. 'We used five different markers with 2, 1.2 and 1.0 mm diameters and a 5 mm length. The two 1.0 mm markers had lengths of 5 mm and 10 mm.'

Additional requirements were:

• Automatic detection (high detection success rate)
• High accuracy (< 1 mm)
• Low imaging dose (1-2 MU)
• Fast analysis (1 s)

a-Si imager uses light more efficiently

The centerpiece of UMCU's study was a small (20 x 20 cm) flat panel imager, mounted on an arm to the gantry of the medical center's Elekta SL20 linear accelerator and controlled by their existing iView™ portal imaging system. In contrast to conventional CCD camera systems, the flat panel imager allows direct digital recording of x-ray images, without the intermediate step of optical or mechanical scanning. The a-Si imager's main components are a light sensitive sensor, a copper plate, and a fluorescent (Lanex) screen along with read-out and driving electronics. The light sensitive sensor has a photodiode array with 256 X 256 pixels (sensors) arranged in rows and columns. Each pixel in the array comprises a light-sensing amorphous-silicon photodiode and a switching thin-film transistor (TFT).

University Medical Center Utrecht

The Lanex screen absorbs x-ray photons and emits photons of visible light. Each sensor on the detector array has a photodiode that detects the converted x-ray photons and changes them into proportional electrical signals, which are then sent to processing electronics.

'The main advantage of a-Si imagers over CCD cameras is that a-Si imagers use light from the detector screen more efficiently, while a CCD camera uses only a small fraction of the light,' Mr. Nederveen explains.

'The a-Si imager doesn't necessarily provide better spatial resolution - because that's dictated by the pixel size of the detector - but it does provide enhanced contrast resolution, which allows us to see the very small gold markers.'

Imaging methodology

In the UMCU study, the five gold markers were arranged on a template and placed on the patient's skin before the irradiation session. All images were obtained under worst case conditions using 18MV photons. 'We used lateral fields and the template was placed on the skin at the beam exit,' Mr. Nederveen explains. 'This procedure results in a marker projection size that is smaller than the projection size for markers inside the body.'

At UMCU, prostate patients are treated with a three-field technique. The lateral beams consist of an open and a wedged part. The images were obtained during the open part of the lateral beam. Just before the open beam came up, the imager was switched on with a frame time of 200 ms. 'We selected the highest possible frame time to maximize the a-Si imager's dynamic range,' he says. 'With a frame time of 200 ms we could get about 20 images per patient. The Elekta SL20 operated at a dose rate of 400-500 MU/min, which means that at a frame time of 200 ms each frame contained about 1.5 MU.'

MEK and edge matching algorithms used

For automatic marker detection, UMCU used a marker extraction kernel (MEK) algorithm that it had developed and used in a previous marker study. 'To put it simply, the MEK is a template with a form that is very comparable to the shape of the marker,' Mr. Nederveen says. 'You move this template over the portal image, and, at every pixel position, you compare the form of the image under the template with the template itself. When you are at the marker position, you will see that the template shape matches very well with the form of the image under the template. So, the MEK allows automatic marker detection in less than one second by determining the discrepancy between the template and what's under the template.'

Aart Nederveen

Essential for automatic position verification is the determination of a fiducial reference point for detection of the gold markers. 'It's one thing to have the coordinate for the marker itself through the MEK, but you also need a reference point related to the detected field edge in the image, commonly the center of gravity of the field,' Mr. Nederveen observes.

The problem is that the dosimetric field edge never coincides exactly with the geometric field edge from treatment planning. Our edge-finding algorithm avoids this mismatch.'

Results

Detection success rate

Compared to UMCU's existing camera-based megavoltage imager, the a-Si imager provides better performance for marker visualization. For example, the detection success rate of a marker of 1.2 mm placed at the beam exit increased from 0.39 to 0.99 (see Table 1).

'It's clear that markers of 1.0 mm diameter can be used for implantation, and the detection success rate for these markers should even be higher than 90 percent in clinical practice,' Mr. Nederveen notes. 'First, because implanted markers will have a larger projection size than markers placed at the beam exit, and we've demonstrated previously that larger projection sizes result in higher success rates. Second, in an actual clinical situation, we would only search for the marker in a restricted area of about 30 x 30 mm in the image. In contrast, the success rates we achieved were based on a search in the total area within the field boundaries.

'Consequently, the use of an a-Si imager for on-line position verification using small markers of 1.0 mm diameter combines the benefits of a clinically acceptable marker size with a clinically acceptable detection success rate.'

Localization accuracy

In a previous study,² Mr. Nederveen evaluated the accuracy by which small markers are found by measuring the distances between small and large markers in the first frames. The distance between a small marker and a large marker of 2.0 mm diameter is a good measure for the localization accuracy of the smaller marker. In Table 2, these distances are given for all small markers averaged over all patients. All localization accuracies were less than 0.3 mm.

For the calculation of distances for each patient, UMCU used the average frame of the first four frames to model the clinical scenario as closely as possible, in which only the first images are used for position verification. These frames contained 1-2 MU in total, as the dose rate is relatively low in the first frames during beam stabilization.

Conclusion

With all requirements for marker implantation met, Mr. Nederveen is confident in the feasibility of automatic marker detection with an a-Si imager. 'Our detection method is very accurate,' he says. 'Adding up the accuracies from marker detection and field edge matching quadratically results in an overall accuracy of 0.4 mm. The overall detection time consists of the time needed for marker detection and the time needed for field edge detection. Both times were found to be smaller than one second, so we believe we can optimize the overall detection time to ±1 second.'

Elekta and UMCU are collaborating to integrate the technique and related algorithm into iViewGT™.

Follow up studies

Study of implanted markers

'Partial boost' dose escalation for prostate cancer

Sophisticated position verification using implanted markers has enabled Mr. Nederveen to initiate a study to evaluate a dose escalation method for the prostate that avoids adjacent normal tissues. 'Giving a high effective dose to the prostate is desired but not routinely done in an effort to avoid exposing the rectum.A dose of more than 70 Gy can cause very high toxicity,' he says. 'Using implanted markers and an on- line position verification method allows us to give a safe boost to the CTV without giving more dose to the rectum.'

Mr. Nederveen sees a continuing need to focus on visualizing and repeatedly verifying the position of clinical targets.

'It's great to have all these sophisticated, evolving techniques like IMRT and inverse dose planning, but I think the most important thing is to verify that the target is where you expect it to be,' he says. 'If you don't have a good standard of position verification, you could actually perform a worse treatment even with the best dose delivery techniques.

'In addition, continued refinements in pre-treatment medical imaging modalities, such as MRI, PET and SPECT, will help enhance radiotherapy precision, as much as IMRT and better dose planning methods enhance treatment delivery.'

References

1. A.J. Nederveen, J.J.W. Lagendijk and P. Hofman. Feasibility of automatic marker detection with an a-Si flat panel imager. Phys. Med. Biol. 46 (2001) 1219-1230.
   
   
2. A.J. Nederveen, J.J.W. Lagendijk and P. Hofman. Detection of fiducial gold markers for automatic on-line megavoltage position verification using a marker extraction kernel (MEK). Int. J. Radiat. Oncol. Biol. Phys. 47 (2000) 1435-1442.