Stephen Breen

Dosimetric comparison of IMRT vs. 3D conformal radiotherapy in the treatment of cancer of the cervical esophagus

BACKGROUND AND PURPOSE Radiotherapy planning for cervical esophageal cancer is challenging. We compared IMRT and 3D conformal radiotherapy (CRT) with respect to conformality of target coverage and normal tissue sparing. MATERIALS AND METHODS We selected five patients with cervical esophagus cancer, who represented the heterogeneity of clinical cases, treated to radical dose and planned with Pinnacle v6.2. Target doses for CRT plans were 50, 60, and 70Gy (single-phase IMRT 56, 63, and 70). We compared PTV coverage by the 95% isodose (PTV(95)), conformality ratio (CR), conformation number (CN), and maximum or mean doses (D(max), D(mean)) to normal structures. RESULTS Median PTV(95) for IMRT plans for PTV70, PTV63, and PTV56 were 97%, 99%, and 98% (CRT 91%, 98%, and 85%). IMRT plans demonstrated lower D(max) to the spinal cord and brainstem (42 and 36Gy) compared to CRT (46 and 39Gy). Median left parotid D(mean) was 35Gy (IMRT) vs. 53Gy (CRT). Median right parotid D(mean) was 35Gy (IMRT) vs. 36Gy (CRT). The median CR50/56Gy was 1.4 (CRT) vs. 1.2 (IMRT), CR70Gy 1.7 (CRT) vs. 1.1 (IMRT). CN50/56 and CN70 values were 0.80 and 0.85 (IMRT) vs. 0.56 and 0.5 (CRT). CONCLUSIONS IMRT provides superior target volume coverage and conformality, with decreased dose to normal structures.

Full orientation invariance and improved feature selectivity of 3D SIFT with application to medical image analysis

This paper presents a comprehensive extension of the Scale Invariant Feature Transform (SIFT), originally introduced in 2D, to volumetric images. While tackling the significant computational efforts required by such multiscale processing of large data volumes, our implementation addresses two important mathematical issues related to the 2D-to-3D extension. It includes efficient steps to filter out extracted point candidates that have low contrast or are poorly localized along edges or ridges. In addition, it achieves, for the first time, full 3D orientation invariance of the descriptors, which is essential for 3D feature matching. An application of this technique is demonstrated to the feature-based automated registration and segmentation of clinical datasets in the context of radiation therapy.

Statistical process control for IMRT dosimetric verification.

Patient-specific measurements are typically used to validate the dosimetry of intensity-modulated radiotherapy (IMRT). To evaluate the dosimetric performance over time of our IMRT process, we have used statistical process control (SPC) concepts to analyze the measurements from 330 head and neck (H&N) treatment plans. The objectives of the present work are to: (i) Review the dosimetric measurements of a large series of consecutive head and neck treatment plans to better understand appropriate dosimetric tolerances; (ii) analyze the results with SPC to develop action levels for measured discrepancies; (iii) develop estimates for the number of measurements that are required to describe IMRT dosimetry in the clinical setting; and (iv) evaluate with SPC a new beam model in our planning system. H&N IMRT cases were planned with the PINNACLE treatment planning system versions 6.2b or 7.6c (Philips Medical Systems, Madison, WI) and treated on Varian (Palo Alto, CA) or Elekta (Crawley, UK) linacs. As part of regular quality assurance, plans were recalculated on a 20-cm-diam cylindrical phantom, and ion chamber measurements were made in high-dose volumes (the PTV with highest dose) and in low-dose volumes (spinal cord organ-at-risk, OR). Differences between the planned and measured doses were recorded as a percentage of the planned dose. Differences were stable over time. Measurements with PINNACLE3 6.2b and Varian linacs showed a mean difference of 0.6% for PTVs (n=149, range, -4.3% to 6.6%), while OR measurements showed a larger systematic discrepancy (mean 4.5%, range -4.5% to 16.3%) that was due to well-known limitations of the MLC model in the earlier version of the planning system. Measurements with PINNACLE3 7.6c and Varian linacs demonstrated a mean difference of 0.2% for PTVs (n=160, range, -3.0%, to 5.0%) and -1.0% for ORs (range -5.8% to 4.4%). The capability index (ratio of specification range to range of the data) was 1.3 for the PTV data, indicating that almost all measurements were within +/-5%. We have used SPC tools to evaluate a new beam model in our planning system to produce a systematic difference of -0.6% for PTVs and 0.4% for ORs, although the number of measurements is smaller (n=25). Analysis of this large series of H&N IMRT measurements demonstrated that our IMRT dosimetry was stable over time and within accepted tolerances. These data provide useful information for assessing alterations to beam models in the planning system. IMRT is enhanced by the addition of statistical process control to traditional quality control procedures.

Auto‐segmentation of normal and target structures in head and neck CT images: A feature‐driven model‐based approach

PURPOSE Intensity modulated radiation therapy (IMRT) allows greater control over dose distribution, which leads to a decrease in radiation related toxicity. IMRT, however, requires precise and accurate delineation of the organs at risk and target volumes. Manual delineation is tedious and suffers from both interobserver and intraobserver variability. State of the art auto-segmentation methods are either atlas-based, model-based or hybrid however, robust fully automated segmentation is often difficult due to the insufficient discriminative information provided by standard medical imaging modalities for certain tissue types. In this paper, the authors present a fully automated hybrid approach which combines deformable registration with the model-based approach to accurately segment normal and target tissues from head and neck CT images. METHODS The segmentation process starts by using an average atlas to reliably identify salient landmarks in the patient image. The relationship between these landmarks and the reference dataset serves to guide a deformable registration algorithm, which allows for a close initialization of a set of organ-specific deformable models in the patient image, ensuring their robust adaptation to the boundaries of the structures. Finally, the models are automatically fine adjusted by our boundary refinement approach which attempts to model the uncertainty in model adaptation using a probabilistic mask. This uncertainty is subsequently resolved by voxel classification based on local low-level organ-specific features. RESULTS To quantitatively evaluate the method, they auto-segment several organs at risk and target tissues from 10 head and neck CT images. They compare the segmentations to the manual delineations outlined by the expert. The evaluation is carried out by estimating two common quantitative measures on 10 datasets: volume overlap fraction or the Dice similarity coefficient (DSC), and a geometrical metric, the median symmetric Hausdorff distance (HD), which is evaluated slice-wise. They achieve an average overlap of 93% for the mandible, 91% for the brainstem, 83% for the parotids, 83% for the submandibular glands, and 74% for the lymph node levels. CONCLUSIONS Our automated segmentation framework is able to segment anatomy in the head and neck region with high accuracy within a clinically-acceptable segmentation time.

Cone-Beam CT Assessment of Interfraction and Intrafraction Setup Error of Two Head-and-Neck Cancer Thermoplastic Masks

PURPOSE To prospectively compare setup error in standard thermoplastic masks and skin-sparing masks (SSMs) modified with low neck cutouts for head-and-neck intensity-modulated radiation therapy (IMRT) patients. METHODS AND MATERIALS Twenty head-and-neck IMRT patients were randomized to be treated in a standard mask (SM) or SSM. Cone-beam computed tomography (CBCT) scans, acquired daily after both initial setup and any repositioning, were used for initial and residual interfraction evaluation, respectively. Weekly, post-IMRT CBCT scans were acquired for intrafraction setup evaluation. The population random (sigma) and systematic (Sigma) errors were compared for SMs and SSMs. Skin toxicity was recorded weekly by use of Radiation Therapy Oncology Group criteria. RESULTS We evaluated 762 CBCT scans in 11 patients randomized to the SM and 9 to the SSM. Initial interfraction sigma was 1.6 mm or less or 1.1 degrees or less for SM and 2.0 mm or less and 0.8 degrees for SSM. Initial interfraction Sigma was 1.0 mm or less or 1.4 degrees or less for SM and 1.1 mm or less or 0.9 degrees or less for SSM. These errors were reduced before IMRT with CBCT image guidance with no significant differences in residual interfraction or intrafraction uncertainties between SMs and SSMs. Intrafraction sigma and Sigma were less than 1 mm and less than 1 degrees for both masks. Less severe skin reactions were observed in the cutout regions of the SSM compared with non-cutout regions. CONCLUSIONS Interfraction and intrafraction setup error is not significantly different for SSMs and conventional masks in head-and-neck radiation therapy. Mask cutouts should be considered for these patients in an effort to reduce skin toxicity.

A Facility for Magnetic Resonance-Guided Radiation Therapy

Magnetic resonance (MR) imaging is routinely employed in the design of radiotherapy (RT) treatment plans for many disease sites. It is evident that tighter integration of MR imaging into the RT process would increase confidence in dose placement and facilitate the integration of new MR imaging information (including anatomical and functional imaging) into the therapy process. To this end, a dedicated MR-guided RT (MRgRT) facility has been created that integrates a state-of-the-art linear accelerator delivery system, high-dose rate brachytherapy afterloader, and superconducting MR scanner to allow MR-based online treatment guidance, adaptive replanning, and response monitoring while maintaining the clinical functionality of the existing delivery systems. This system is housed within a dedicated MRgRT suite and operates in a coordinated fashion to assure safe and efficient MRgRT treatments.

Biomechanical-based image registration for head and neck radiation treatment*

Deformable image registration of four head and neck cancer patients has been conducted using a biomechanical-based model. Patient-specific 3D finite element models have been developed using CT and cone-beam CT image data of the planning and a radiation treatment session. The model consists of seven vertebrae (C1 to C7), mandible, larynx, left and right parotid glands, tumor and body. Different combinations of boundary conditions are applied in the model in order to find the configuration with a minimum registration error. Each vertebra in the planning session is individually aligned with its correspondence in the treatment session. Rigid alignment is used for each individual vertebra and the mandible since no deformation is expected in the bones. In addition, the effect of morphological differences in the external body between the two image sessions is investigated. The accuracy of the registration is evaluated using the tumor and both parotid glands by comparing the calculated Dice similarity index of these structures following deformation in relation to their true surface defined in the image of the second session. The registration is improved when the vertebrae and mandible are aligned in the two sessions with the highest average Dice index of 0.86 ± 0.08, 0.84 ± 0.11 and 0.89 ± 0.04 for the tumor, left and right parotid glands, respectively. The accuracy of the center of mass location of tumor and parotid glands is also improved by deformable image registration where the errors in the tumor and parotid glands decrease from 4.0 ± 1.1, 3.4 ± 1.5 and 3.8 ± 0.9 mm using rigid registration to 2.3 ± 1.0, 2.5 ± 0.8 and 2.0 ± 0.9 mm in the deformable image registration when alignment of vertebrae and mandible is conducted in addition to the surface projection of the body.

Quantitative PET Comparing Gated with Nongated Acquisitions Using a NEMA Phantom with Respiratory-Simulated Motion

This study evaluated the use of gated versus nongated PET acquisitions for absolute quantification of radioisotope concentration (RC) in a respiratory motion–simulated moving phantom filled with radioactive spheres and background for both 2-dimensional (2D) and 3-dimensional (3D) acquisitions. Methods: An image-quality phantom with all 6 spheres filled with the same 18F RC (range, 19–62 kBq/mL) was scanned with PET/CT at rest and in motion with and without gating. The background was filled with 18F solution to yield sphere-to-background ratios of approximately 5, 10, 15, and 20 to 1. Both 2D and 3D acquisitions were used for all combinations. Respiratory motion was simulated by using a motor-driven plastic platform to move the phantom periodically with a displacement of 2 cm and a cycle time of 5.8 s. For gated acquisitions, the phantom was tracked using a real-time position management system. Images were reconstructed, and regions of interest with the same sizes as the actual spheres were manually placed on axial slices to determine maximum and mean pixel RC. A threshold method (70% and 94% for 2D and 3D modes) was also used to determine a mean voxel RC. All values were compared with the expected RC; percentage differences were calculated for each sphere. To reduce partial-volume effects, only data for the 4 largest spheres were analyzed. Results: The mean pixel method was the only method with linear responses for all 3 scan types, enabling direct comparisons. The ranges of RC percentage differences were underestimated for all scan types (using the mean pixel method). The overall mean percentage differences were 37, 49, and 41 in 2D mode and 40, 51, and 41 in 3D mode for static, nongated, and gated acquisitions, respectively. Gated acquisitions improved quantification (by reducing underestimation) over nongated acquisitions by 8% and 10% for 2D and 3D modes. Conclusion: In the presence of motion, the use of gated PET acquisitions appears to improve quantification accuracy over nongated acquisitions, almost restoring the results to those observed when the phantom is static.

Improving Observer Variability in Target Delineation for Gastro-oesophageal Cancer—the Role of 18Ffluoro-2-deoxy-d-glucose Positron Emission Tomography/Computed Tomography

AIM To evaluate the effect of the addition of fused positron emission tomography-computed tomography (PET-CT) imaging vs computed tomography alone in the identification of the gross tumour volume (GTV) in patients with gastro-oesophageal carcinoma. MATERIALS AND METHODS Ten patients with gastro-oesophageal cancer referred for radiation therapy underwent both (18F)fluoro-2-deoxy-d-glucose-PET (FDG-PET) and computed tomography in the treatment position. Image sets were anonymised and co-registered. Six radiation oncologists independently defined the GTV, first using the computed tomography data alone supplemented by standardised clinical and diagnostic imaging information, and second, using co-registered computed tomography and FDG-PET data (PET-CT). The standard deviation for both GTV length and volume (excluding involved lymph nodes) was taken as a measurement of inter-observer and intra-observer variability. Computer software that calculates volume overlap between contours was also used to generate an observer agreement index to compare intra- and inter-observer variability. RESULTS The addition of FDG-PET imaging decreased the median standard deviation for tumour length from 10 mm (range 8.1-33.3, mean 12.4 mm) for computed tomography alone to 8mm (range 4.4-18.1, mean 8.1 mm) for PET-CT (P = 0.02). Eight of the 10 patients showed an increase in volume of overlap between observers with the addition of FDG-PET imaging to the contouring process (P = 0.05). The average observer agreement index in PET-CT was 72.7% compared with 69.1% when using computed tomography alone. There was significantly less intra-observer variability in all measures when PET-CT was used. The median standard deviation in length improved from 5.3 to 1.8 mm, the median standard deviation in volume improved from 4.5 to 3 cm3 and the median observer agreement index improved from 76.2 to 78.7% when computed tomography alone was compared with PET-CT. The corresponding P values were 0.001, 0.033 and 0.022, respectively. CONCLUSIONS The addition of FDG-PET to computed tomography-based planning for the identification of primary tumour GTV in patients with gastro-oesophageal carcinoma decreases both inter-observer and intra-observer variability.

Improving superficial target delineation in radiation therapy with endoscopic tracking and registration

PURPOSE Target delineation within volumetric imaging is a critical step in the planning process of intensity modulated radiation therapy. In endoluminal cancers, endoscopy often reveals superficial areas of visible disease beyond what is seen on volumetric imaging. Quantitatively relating these findings to the volumetric imaging is prone to human error during the recall and contouring of the target. We have developed a method to improve target delineation in the radiation therapy planning process by quantitatively registering endoscopic findings contours traced on endoscopic images to volumetric imaging. METHODS Using electromagnetic sensors embedded in an endoscope, 2D endoscopic images were registered to computed tomography (CT) volumetric images by tracking the position and orientation of the endoscope relative to a CT image set. Regions-of-interest (ROI) in the 2D endoscopic view were delineated. A mesh created within the boundary of the ROI was projected onto the 3D image data, registering the ROI with the volumetric image. This 3D ROI was exported to clinical radiation treatment planning software. The precision and accuracy of the procedure was tested on two solid phantoms with superficial markings visible on both endoscopy and CT images. The first phantom was T-shaped tube with X-marks etched on the interior. The second phantom was an anatomically correct skull phantom with a phantom superficial lesion placed on the pharyngeal surface. Markings were contoured on the endoscope images and compared with contours delineated in the treatment planning system based on the CT images. Clinical feasibility was tested on three patients with early stage glottic cancer. Image-based rendering using manually identified landmarks was used to improve the registration. RESULTS Using the T-shaped phantom with X-markings, the 2D to 3D registration accuracy was 1.5-3.5 mm, depending on the endoscope position relative to the markings. Intraobserver standard variation was 0.5 mm. Rotational accuracy was within 2°. Using the skull phantom, registration accuracy was assessed by calculating the average surface minimum distance between the endoscopy and treatment planning contours. The average surface distance was 0.92 mm with 93% of all points in the 2D-endoscopy ROI within 1.5 mm of any point within the ROI contoured in the treatment planning software. This accuracy is limited by the CT imaging resolution and the electromagnetic (EM) sensor accuracy. The clinical testing demonstrated that endoscopic contouring is feasible. With registration based on em tracking only, accuracy was 5.6-8.4 mm. Image-based registration reduced this error to less than 3.5 mm and enabled endoscopic contouring in all cases. CONCLUSIONS Registration of contours generated on 2D endoscopic images to 3D planning space is feasible, with accuracy smaller than typical set-up margins. Used in addition to standard 3D contouring methods in radiation planning, the technology may improve gross tumour volume (GTV) delineation for superficial tumors in luminal sites that are only visible in endoscopy.