Automatic Treatment Planning for Multi-focal Dynamic Conformal Arc GRID Therapy for Late-Stage Lung Cancer: A Feasibility Study

PURPOSE/OBJECTIVE(S)

Palliative management of large, symptomatic pulmonary lesions, either as primary lung cancers or metastases, can be challenging due to need to balance effective radiation doses for cytoreduction with safety. Spatially Fractionated Radiation Therapy (SFRT), or GRID Therapy, is an emerging technique, which delivers ablative doses of radiotherapy to small, selected areas of tumor, while sparing organs-at-risk (OARs), and has been shown to debulk large lesions in preliminary studies. Conventionally, an alloy GRID block is manufactured to deliver GRID therapy. However, this delivery technique poses a challenge due to need for block, and dosimetrically when the tumor is deep-seated as excess dose may be delivered to OARs, such as skin and chest wall. This study aims to develop a fast, automatic planning solution using multi-focal dynamic conformal arcs (DCA) on modern Linear Accelerator.

MATERIALS/METHODS

One late-stage lung cancer patient with simulated sphere target grid was included in this study. The sphere targets are 1.5cm in diameter and 4.3cm spacing. Four co-planar full arcs were used for optimization. The problem is formalized as finding optimal multi-leaf collimator (MLC) sequencing to cover N targets with K control points (CPs) for each arc. The state of each target's MLC opening at each CP is binary. In order to solve this NP-hard problem, the optimal solution was approximated by eliminating projection collision at each CP. MLC motion continuity and maximum speed were included in the cost function to ensure deliverability. The optimization started with randomized initial CP apertures, followed by solving state-transition equations for following CPs. Two grid arrays (9 and 10 targets respectively) were tested for plan quality. For each grid of target, the arc collimator angle was planned with 0 and 30 degrees for comparison. Prescription was 20 Gy per fraction. Monte Carlo simulation dose engine from matRad toolkit was used for dose calculation. Key dosimetric endpoints including target mean dose, D5%(Gy) and D95%(Gy), were reported.

RESULTS

Average calculation time on the AMD Ryzen 5 5600 × 6-Core 3.7GHz CPU and 32GB RAM platform varied from 31 to 44 minutes. One zero-degree collimator and one thirty-degree collimator were generated for each target array. For nine-target array, mean target dose from both plans ranged from 23.41 to 26.55 Gy, while D5%(Gy) and D95%(Gy) ranged from 25.45 to 30.16 Gy, and 20.00 to 22.21 Gy, respectively. For ten-target array, the range of target mean, D5%(Gy) and D95%(Gy) were 23.82 to 28.74 Gy, 26.50 to 33.11 Gy, and 20.00 to 22.49 Gy.

CONCLUSION

A fast, automatic planning solution for multi-focal DCA GRID therapy was developed. It provides clinically feasible plans with high efficiency for small target arrays for the late-stage cancer patient. The implementation provides excellent coverage for deep-seated tumors where alloy grid solution could fail to meet coverage objectives. Additional patients are needed in the future to further refine the technique.

A Radiogenomic Machine Learning Model for Glioblastoma Post-Resection Overall Survival Group Prediction

PURPOSE/OBJECTIVE(S)

Novel methods are needed to better predict outcomes and tailor personalized therapies for patients with glioblastoma (GBM). To improve post-resection survival prediction in GBM patients, we developed a machine learning model that utilized 1) radiomic computational biomarkers from pre-resection multi-parametric MRI (mp-MRI) images, and 2) clinical information including tumor molecular features and extent of surgical resection.

MATERIALS/METHODS

A cohort of 406 GBM patients treated with surgical resection was studied. Each patient received a pre-resection mp-MRI that included T1, contrast-enhanced T1 (T1-ce), T2, and FLAIR sequences. Three tumor subregions, i.e., enhanced tumor, tumor core, and whole tumor, were segmented with radiologists' correction. Based on survival outcomes in the literature, the cohort was categorized into three survival groups: group A (<9 mos, n = 148), B (9-21 mos, n = 181), and C (21+ mos, n = 77). We first extracted 88 radiomic features from each tumor subregion on each MR volume, and Z-score normalization was adopted. Three other patient-specific factors, including age, resection status (subtotal versus gross total), and IDH1 status, were concatenated with the radiomic features as a synthesized patient-specific feature vector. We then designed a two-step machine learning model: using the patient-specific feature vectors, the model 1) identified patients in group A with the shortest predicted survival using a balanced random forest (BRF) classifier, and 2) used a 2nd BRF classifier to segregate the remaining patients into groups B and C. For model training, a 7:3 training/test sample ratio was adopted, and 100 model versions were acquired through random validation sample assignments to study model robustness. Sensitivity, specificity, and accuracy results of each group were calculated, and an overall ROC was generated to represent the model's overall performance.

RESULTS

The model demonstrated acceptable prediction performance with an ROC AUC value at 0.68. Individually, the model achieved good prediction accuracies for short and long-term survival prediction (Groups A and C), though we observed relatively limited accuracy in medium survival prediction (Group B). Model sensitivity in group A was promising, but was limited in the remaining two groups. Feature weight analysis showed that radiomic features from the enhanced tumor subregion were the leading variables in the BRF classifier.

CONCLUSION

The developed radiogenomic machine learning model predicts GBM post-resection overall survival outcomes. Future work is necessary to further improve model sensitivity for patients with medium and long-term predicted survival.

Evaluation of the effect of respiratory and anatomical variables on a Fourier technique for markerless, self-sorted 4D-CBCT

A novel technique based on Fourier transform theory has been developed that directly extracts respiratory information from projections without the use of external surrogates. While the feasibility has been demonstrated with three patients, a more extensive validation is necessary. Therefore, the purpose of this work is to investigate the effects of a variety of respiratory and anatomical scenarios on the performance of the technique with the 4D digital extended cardiac torso phantom. FT-phase and FT-magnitude methods were each applied to identify peak-inspiration projections and quantitatively compared to the gold standard of visual identification. Both methods proved to be robust across the studied scenarios with average differences in respiratory phase <10% and percentage of projections assigned within 10% of the gold standard >90%, when incorporating minor modifications to region-of-interest (ROI) selection and/or low-frequency location for select cases of DA and lung percentage in the field of view of the projection. Nevertheless, in the instance where one method initially faltered, the other method prevailed and successfully identified peak-inspiration projections. This is promising because it suggests that the two methods provide complementary information to each other. To ensure appropriate clinical adaptation of markerless, self-sorted four-dimensional cone-beam CT (4D-CBCT), perhaps an optimal integration of the two methods can be developed.

The effect of motion on IMRT - looking at interplay with 3D measurements

Six base of skull IMRT treatment plans were delivered to 3D dosimeters within the RPC Head and Neck Phantom for QA verification. Isotropic 2mm 3D data was obtained using the DLOS-PRESAGE system and compared to an Eclipse (Varian) treatment plan. Normalized Dose Distribution pass rates were obtained for a number of criteria. High quality 3D dosimetry data was observed from the DLOS system, illustrated here through colormaps, isodose lines, profiles, and NDD 3D maps. Excellent agreement with the planned dose distributions was also observed with NDD analysis revealing > 90% NDD pass rates [3%, 2mm], noise < 0.5%. This paper focuses on a detailed exploration of the quality and use of 3D dosimetry data obtained with the DLOS-PRESAGE system.

Stereotactic Radiotherapy for Malignancies Involving the Trigeminal and Facial Nerves

Involvement of a cranial nerve caries a poor prognosis for many malignancies. Recurrent or residual disease in the trigeminal or facial nerve after primary therapy poses a challenge due to the location of the nerve in the skull base, the proximity to the brain, brainstem, cavernous sinus, and optic apparatus and the resulting complex geometry. Surgical resection caries a high risk of morbidity and is often not an option for these patients. Stereotactic radiosurgery and radiotherapy are potential treatment options for patients with cancer involving the trigeminal or facial nerve. These techniques can deliver high doses of radiation to complex volumes while sparing adjacent critical structures. In the current study, seven cases of cancer involving the trigeminal or facial nerve are presented. These patients had unresectable recurrent or residual disease after definitive local therapy. Each patient was treated with stereotactic radiation therapy using a linear accelerator based system. A multidisciplinary approach including neuroradiology and surgical oncology was used to delineate target volumes. Treatment was well tolerated with no acute grade 3 or higher toxicity. One patient who was reirradiated experienced cerebral radionecrosis with mild symptoms. Four of the seven patients treated had no evidence of disease after a median follow up of 12 months (range 2–24 months). A dosimetric analysis was performed to compare intensity modulated fractionated stereotactic radiation therapy (IM-FSRT) to a 3D conformal technique. The dose to 90% (D90) of the brainstem was lower with the IM-FSRT plan by a mean of 13.5 Gy. The D95 to the ipsilateral optic nerve was also reduced with IM-FSRT by 12.2 Gy and the D95 for the optic chiasm was lower with FSRT by 16.3 Gy. Treatment of malignancies involving a cranial nerve requires a multidisciplinary approach. Use of an IM-FSRT technique with a micro-multileaf collimator resulted in a lower dose to the brainstem, optic nerves and chiasm for each case examined.

A methodology for selecting the beam arrangement to reduce the intensity-modulated radiation therapy (IMRT) dose to the SPECT-defined functioning lung

Macroaggregated albumin single-photon emission computed tomography (MAA-SPECT) provides a map of the spatial distribution of lung perfusion. Our previous work developed a methodology to use SPECT guidance to reduce the dose to the functional lung in IMRT planning. This study aims to investigate the role of beam arrangement on both low and high doses in the functional lung. In our previous work, nine-beam IMRT plans were generated with and without SPECT guidance and compared for five patients. For the current study, the dose-function histogram (DFH) contribution for each of the nine beams for each patient was calculated. Four beams were chosen based on orientation and DFH contributions to create a SPECT-guided plan that spared the functional lung and maintained target coverage. Four-beam SPECT-guided IMRT plans reduced the F(20) and F(30) values by (16.5 +/- 6.8)% and (6.1 +/- 9.2)%, respectively, when compared to nine-beam conventional IMRT plans. Moreover, the SPECT-4F Plan reduces F(5) and F(13) for all patients by (11.0 +/- 8.2)% and (6.1 +/- 3.6)%, respectively, compared to the SPECT Plan. Using fewer beams in IMRT planning may reduce the amount of functional lung that receives 5 and 13 Gy, a factor that has recently been associated with radiation pneumonitis.

Tradeoffs of integrating real-time tracking into IGRT for prostate cancer treatment

This study investigated the integration of the Calypso real-time tracking system, based on implanted ferromagnetic transponders and a detector array, into the current process for image-guided radiation treatment (IGRT) of prostate cancer at our institution. The current IGRT process includes magnetic resonance imaging (MRI) for prostate delineation, CT simulation for treatment planning, daily on-board kV and CBCT imaging for target alignment, and MRI/MRS for post-treatment assessment. This study assesses (1) magnetic-field-induced displacement and radio-frequency (RF)-induced heating of transponders during MRI at 1.5 T and 3 T, and (2) image artifacts caused by transponders and the detector array in phantom and patient cases with the different imaging systems. A tissue-equivalent phantom mimicking prostate tissue stiffness was constructed and implanted with three operational transponders prior to phantom solidification. The measurements show that the Calypso system is safe with all the imaging systems. Transponder position displacements due to the MR field are minimal (<1.0 mm) for both 1.5 T and 3 T MRI scanners, and the temperature variation due to MRI RF heating is <0.2 degrees C. The visibility of transponders and bony anatomy was not affected on the OBI kV and CT images. Image quality degradation caused by the detector antenna array is observed in the CBCT image. Image artifacts are most significant with the gradient echo sequence in the MR images, producing null signals surrounding the transponders with radii approximately 1.5 cm and length approximately 4 cm. Thus, Calypso transponders can preclude the use of MRI/MRS in post-treatment assessment. Modifications of the clinical flow are required to accommodate and minimize the substantial MRI artifacts induced by the Calypso transponders.

A three-dimensional deformable model for segmentation of human prostate from ultrasound images

Segmentation of human prostate from ultrasound (US) images is a crucial step in radiation therapy, especially in real-time planning for US image-guided prostate seed implant. This step is critical to determine the radioactive seed placement and to ensure the adequate dose coverage of prostate. However, due to the low contrast of prostate and very low signal-to-noise ratio in US images, this task remains as an obstacle. The manual segmentation of this object is time consuming and highly subjective. In this work, we have proposed a three-dimensional (3D) deformable surface model for automatic segmentation of prostate. The model has a discrete structure made from a set of vertices in the 3D space that form triangle facets. The model converges from an initial shape to its equilibrium iteratively, by a weighted sum of the internal and external forces. Internal forces are based on the local curvature of the surface and external forces are extracted from the volumetric image data by applying an appropriate edge filter. We have also developed a method for initialization of the model from a few initial contours that are drawn on different slices. During the deformation, a resampling procedure is used to maintain the resolution of the model. The entire model is applied in a multiscale scheme, which increases the robustness and speed, and guarantees a better convergence. The model is tested on real clinical data and initial results are very promising.

Adaptive portal CT reconstruction: a simulation study

In radiotherapy, radiation treatment beams contain valuable information for patient setup verification. These beams may be used for portal CT reconstruction. However, direct use of the beam data for reconstruction may yield inadequate CT images simply because these beams cover only a part of the patient body. In this study, we use the treatment beams in addition to a set of regular CT projection beams to reconstruct a locally enhanced portal CT image. This approach is called adaptive portal CT reconstruction. A computer simulation demonstrated the advantages of the approach. The image reconstruction was carried out by the multilevel scheme algebraic reconstruction technique. Results indicated that the image quality of adaptive portal CT reconstruction is equivalent to that obtained from a full set of projections. This proposed technique should be not only valuable for three-dimensional radiotherapy verification, but also applicable to diagnostic CT imaging.

Optimization of inverse treatment planning using a fuzzy weight function

A fuzzy approach has been applied to inverse treatment planning optimization in radiation therapy. The proposed inverse-planning algorithm optimizes both the intensity-modulated beam (IMB) and the normal tissue prescription. In the IMB optimization, we developed a fast-monotonic-descent (FMD) method that has the property of fast and monotonic convergence to the minimum for a constrained quadratic objective function. In addition, a fuzzy weight function is employed to express the vague knowledge about the importance of matching the calculated dose to the prescribed dose in the normal tissue. Then, a validity function is established to optimize the normal tissue prescription. The performance of this new fuzzy prescription algorithm has been compared to that based on hard prescription methods for two treatment geometries. The FMD method presented here both provides a full-analytical solution to the optimization of intensity-modulated beams, and guarantees fast and monotonic convergence to the minimum. It has been shown that the fuzzy inverse planning technique is capable of achieving an optimal balance between the objective of matching the calculated dose to the prescribed dose for the target volume and the objective of minimizing the normal tissue dose.

Structures of yeast vesicle trafficking proteins

In protein transport between organelles, interactions of v- and t-SNARE proteins are required for fusion of protein-containing vesicles with appropriate target compartments. Mammalian SNARE proteins have been observed to interact with NSF and SNAP, and yeast SNAREs with yeast homologues of NSF and SNAP proteins. This observation led to the hypothesis that, despite low sequence homology, SNARE proteins are structurally similar among eukaryotes. SNARE proteins can be classified into two groups depending on whether they interact with SNARE binding partners via conserved glutamine (Q-SNAREs) or arginine (R-SNAREs). Much of the published structural data available is for SNAREs involved in exocytosis (either in yeast or synaptic vesicles). This paper describes circular dichroism, Fourier transform infrared spectroscopy, and dynamic light scattering data for a set of yeast v- and t-SNARE proteins, Vti1p and Pep12p, that are Q-SNAREs involved in intracellular trafficking. Our results suggest that the secondary structure of Vti1p is highly alpha-helical and that Vti1p forms multimers under a variety of solution conditions. In these respects, Vti1p appears to be distinct from R-SNARE proteins characterized previously. The alpha-helicity of Vti1p is similar to that of Q-SNARE proteins characterized previously. Pep12p, a Q-SNARE, is highly alpha-helical. It is distinct from other Q-SNAREs in that it forms dimers under many of the solution conditions tested in our experiments. The results presented in this paper are among the first to suggest heterogeneity in the functioning of SNARE complexes.

Treatment field shape verification using elliptic Fourier transform

An automated field shape correlation technique based on elliptic Fourier transform (EFT) is developed to verify the radiation treatment field in digital portal images. In this method, the edge of the treatment field is initially extracted from the portal image and is then approximated by a polygon. The polygon is further represented with elliptic Fourier coefficients. The invariants to shift, rotation, and scale are computed from the elliptic Fourier coefficients to characterize the genuine shape feature and are used to match the reference treatment field. Invariants calculated from both test and reference field shapes are compared to determine the similarity between two treatment fields. The proposed procedure uses the first approved field shape as the reference for automated comparison with subsequent portal images. This technique not only verifies the shape of each portal field but also provides information about relative shift, rotation, and scale. A set of generic shapes is simulated to test the robustness of the algorithm and to determine the parameters used in the decision procedure. Experimental results on the simulated shapes show that this method can detect shape distortions of 2% in area and the standard deviations are 0 for shifting, 0.24 degrees for rotation, and 0.0031 for scaling. Preliminary tests on clinical portal images indicated that this technique is potentially useful for automated real-time portal verification.

MR image-guided portal verification for brain treatment field

PURPOSE

To investigate a method for the generation of digitally reconstructed radiographs directly from MR images (DRR-MRI) to guide a computerized portal verification procedure.

METHODS AND MATERIALS

Several major steps were developed to perform an MR image-guided portal verification procedure. Initially, a wavelet-based multiresolution adaptive thresholding method was used to segment the skin slice-by-slice in MR brain axial images. Some selected anatomical structures, such as target volume and critical organs, were then manually identified and were reassigned to relatively higher intensities. Interslice information was interpolated with a directional method to achieve comparable display resolution in three dimensions. Next, a ray-tracing method was used to generate a DRR-MRI image at the planned treatment position, and the ray tracing was simply performed on summation of voxels along the ray. The skin and its relative positions were also projected to the DRR-MRI and were used to guide the search of similar features in the portal image. A Canny edge detector was used to enhance the brain contour in both portal and simulation images. The skin in the brain portal image was then extracted using a knowledge-based searching technique. Finally, a Chamfer matching technique was used to correlate features between DRR-MRI and portal image.

RESULTS

The MR image-guided portal verification method was evaluated using a brain phantom case and a clinical patient case. Both DRR-CT and DRR-MRI were generated using CT and MR phantom images with the same beam orientation and then compared. The matching result indicated that the maximum deviation of internal structures was less than 1 mm. The segmented results for brain MR slice images indicated that a wavelet-based image segmentation technique provided a reasonable estimation for the brain skin. For the clinical patient case with a given portal field, the MR image-guided verification method provided an excellent match between features in both DRR-MRI and portal image. Moreover, target volume could be accurately visualized in the DRR-MRI and mapped over to the corresponding portal image for treatment verification. The accuracy of DRR-MRI was also examined by comparing it to the corresponding simulation image. The matching results indicated that the maximum deviation of anatomical features was less than 2.5 mm.

CONCLUSION

A method for MR image-guided portal verification of brain treatment field was developed. Although the radiographic appearance in the DRR-MRI is different from that in the portal image, DRR-MRI provides essential anatomical features (landmarks and target volume) as well as their relative locations to be used as references for computerized portal verification.

Oncologic image compression using both wavelet and masking techniques

A new algorithm has been developed to compress oncologic images using both wavelet transform and field masking methods. A compactly supported wavelet transform is used to decompose the original image into high- and low-frequency subband images. The region-of-interest (ROI) inside an image, such as an irradiated field in an electronic portal image, is identified using an image segmentation technique and is then used to generate a mask. The wavelet transform coefficients outside the mask region are then ignored so that these coefficients can be efficiently coded to minimize the image redundancy. In this study, an adaptive uniform scalar quantization method and Huffman coding with a fixed code book are employed in subsequent compression procedures. Three types of typical oncologic images are tested for compression using this new algorithm: CT, MRI, and electronic portal images with 256 x 256 matrix size and 8-bit gray levels. Peak signal-to-noise ratio (PSNR) is used to evaluate the quality of reconstructed image. Effects of masking and image quality on compression ratio are illustrated. Compression ratios obtained using wavelet transform with and without masking for the same PSNR are compared for all types of images. The addition of masking shows an increase of compression ratio by a factor of greater than 1.5. The effect of masking on the compression ratio depends on image type and anatomical site. A compression ratio of greater than 5 can be achieved for a lossless compression of various oncologic images with respect to the region inside the mask. Examples of reconstructed images with compression ratio greater than 50 are shown.

The effect of detector size to the broadening of the penumbra--a computer simulated study

The effect of detector size to the broadening of the measured beam penumbra has been a subject of numerous studies. Based on measured data, linear and quadratic curves have been proposed to describe the relationship between the measured penumbra width, between 10%-90% and 20%-80% intensities, and the detector size. Extrapolations of these curves to zero detector size also suggest that the inherent penumbras can be deduced. However, due to experimental noise, especially when a small ionization chamber is used, and the inherent penumbra is not known, it is difficult to discern the superiority of either model. In this study, one dimensional convolution using a thin circular disk shape detector was employed to analyze the effect of detector size to the broadening of the penumbra. A set of beams with different inherent penumbra widths, ranging from 0 to 350 in arbitrary unit, was first generated. Each beam was then convoluted with the response function of the detectors with different sizes from 10 to 280, in arbitrary unit. The result is an output signal with penumbra that is wider than the inherent penumbra. The plots of penumbra widths to detector radii are a family of concave parabolic curves with different inherent penumbra widths. The concave portions of the curve represent results from scanning with detectors equal to or smaller than the penumbra width, and the linear portions represent results from scanning with detectors much larger than the penumbra. The curves are nothing but different scales of one curve. The extrapolations of the curves to the penumbra axis when the detector radius approaches zero give the deduced inherent penumbra widths. The deduced inherent penumbra widths approximate the inherent penumbra widths satisfactorily. From the graphs provided, the inherent penumbra can be deduced using the detector radius and the measured penumbra width.

An observer study for direct comparison of clinical efficacy of electronic to film portal images

PURPOSE

To directly compare clinical efficacy of electronic to film portal images.

METHODS AND MATERIALS

An observer study was designed to compare clinical efficacy of electronic to film portal images acquired using a liquid matrix ion-chamber electronic portal imaging device and a conventional metal screen/film system. Both images were acquired simultaneously for each treatment port and the electronic portal images were printed on gray-level thermal paper. Four radiation oncologists served as observers and evaluated a total of 44 sets of images for four different treatment sites: lung, pelvis, brain, and head/neck. Each set of images included a simulation image, a double-exposure portal film, and video paper prints of electronic portal images. Eight to nine anatomical landmarks were selected from each treatment site. Each observer was asked to rate each landmark in terms of its clinical visibility and to rate the ease of making the pertinent verification decision in the corresponding electronic and film portal images with the aid of the simulation image.

RESULTS

Ratings for the visibility of landmarks and for the verification decision of treatment ports were similar for electronic and film images for most landmarks. However, vertebral bodies and several landmarks in the pelvis such as the acetabulum and public symphysis were more visible in the portal film images than in the electronic portal images.

CONCLUSION

The visibility of landmarks in electronic portal images is comparable to that in film portal images. Verification of treatment ports based only on electronic portal images acquired using an electronic portal imaging device is generally achievable.

A technique of automating compensator design for lung inhomogeneity correction using an electron portal imaging device

A technique of automating compensator design for lung inhomogeneity correction using an electron portal imaging device (EPID) has been investigated. This technique utilizes exit-radiation information as detected by an EPID to determine the thickness of the compensator desired. In this particular study, the compensator thickness is determined to provide a uniform gray-level distribution (related to uniform exit-dose distribution) in the region of the portal image to be compensated. Initially, a compensation characteristic curve, which relates the compensator thickness to the pixel value of the electronic portal image, is measured for both the Lead and Lipowitz compensator materials and a 6-MV photon beam. Then, a chest-treatment field is simulated using an anthropomorphic phantom. Based on the analysis of the profile (gray-level distribution) across the lung and mediastinum regions in the electronic portal image, the average of pixel values within the mediastinum region is selected as the matching level and the regions to be compensated are determined. With the aid of the predetermined compensation characteristic curve and proper distance scaling, the compensator thickness at each pixel location is automatically calculated at the block tray level to correct lung inhomogeneity. In a simple test using a single anterioposterior (AP) chest field, the compensated profile in the electronic portal image presents a uniform gray-level distribution (related to uniform exit dose) compared to the uncompensated profile.(ABSTRACT TRUNCATED AT 250 WORDS)

Input/output characteristics of a matrix ion-chamber electronic portal imaging device

The input/output characteristics of a matrix liquid ion-chamber electronic portal imaging device (EPID) are investigated to elucidate the imaging properties of EPIDs. The radiation input to the detector, represented by dose rate, and the pixel value output from the device are related by a characteristic curve. Various incident radiation intensities are obtained by changing the source-to-detector distance (SDD). For each incident radiation intensity, an electronic portal image is obtained using a field size of 5 x 5 cm2. The output pixel value of the EPID is represented by the average pixel value of a region of interest of 9 x 9 pixels centered at a selected point. The effects of various accelerator settings, such as the repetition-rate setting and photon energy, gantry angle, field size, SDD, and acquisition mode of the EPID on characteristic curves are investigated at the central axis. The off-axis response of the detector is also examined. The derivative of the pixel value with respect to the input dose rate is used to analyze the detector contrast. Results indicate that the output pixel value is not a linear function of the incident radiation intensity. The detector contrast is comparable between photon energies of 10 and 6 MV and increases at low dose rates. The response of the imaging device varies substantially with acquisition mode, but is less sensitive to the SDD used for calibration. Characteristic curves are consistent for different gantry angles at the central axis and with the off-axis locations when the gantry angle is used for imaging and calibration, but vary with off-axis locations when the gantry angle is not at the calibration direction. Characteristic curves are also found to vary with different field sizes, but are similar in shape.

Computerized detection of masses in digital mammograms: automated alignment of breast images and its effect on bilateral-subtraction technique

An automated technique for the alignment of right and left breast images has been developed for use in the computerized analysis of bilateral breast images. In this technique, the breast region is first identified in each digital mammogram by use of histogram analysis and morphological filtering operations. The anterior portions of the tracked breast border and computer-identified nipple positions are selected as landmarks for use in image registration. The paired right and left breast images, either from mediolateral oblique or craniocaudal views, are then registered relative to each other by use of a least-squares matching method. This automated alignment technique has been applied to our computerized detection scheme that employs a nonlinear bilateral-subtraction method for the initial identification of possible masses. The effectiveness of using bilateral subtraction in identifying asymmetries between corresponding right and left breast images is examined by comparing detection performances obtained with various computer-simulated misalignments of 40 pairs of clinical mammograms. Based on free-response receiver operating characteristic and regression analyses, the detection performance obtained with the automated alignment technique was found to be higher than that obtained with simulated misalignments. Detection performance decreased gradually as the amount of simulated misalignment increased. These results indicate that automatic alignment of breast images is possible and that mass-detection performance appears to improve with the inclusion of asymmetric anatomic information but is not sensitive to slight misalignment.

Computerized detection of masses in digital mammograms: investigation of feature-analysis techniques

Mammographic screening of asymptomatic women has shown effectiveness in the reduction of breast cancer mortality. We are developing a computerized scheme for the detection of mammographic masses as an aid to radiologists in mammographic screening programs. Possible masses on digitized screen/film mammograms are initially identified using a nonlinear bilateral-subtraction technique, which is based on asymmetric density patterns occurring in corresponding portions of right and left mammograms. In this study, we analyze the characteristics of actual masses and nonmass detections to develop feature-analysis techniques with which to reduce the number of nonmass (ie, false-positive) detections. These feature-analysis techniques involve (1) the extraction of various features (such as area, contrast, circularity and border-distance based on the density and geometric information of masses in both processed, and original breast images), and (2) tests of the extracted features to reduce nonmass detections. Cumulative histograms of both actual-mass detections and nonmass detections are used to characterize extracted features and to determine the cutoff values used in the feature tests. The effectiveness of the feature-analysis techniques is evaluated in combination with the computerized detection scheme that uses the nonlinear bilateral-subtraction technique using free-response receiver operating characteristic analysis and 77 patient cases (308 mammograms). Results show that the feature-analysis techniques effectively improve the performance of the computerized detection scheme: about 35% false-positive detections were eliminated without loss in sensitivity when the feature-analysis techniques were used.

Effect of case selection on the performance of computer-aided detection schemes

The choice of clinical cases used to train and test a computer-aided diagnosis (CAD) scheme can affect the test results (i.e., error rate). In this study, we deliberately modified the components of our testing database to study the effects of this modification on measured performance. Using a computerized scheme for the automated detection of breast masses from mammograms, it was found that the sensitivity of the scheme ranged between 26% and 100% (at a false positive rate of 1.0 per image) depending on the cases used to test the scheme. Even a 20% change in the cases comprising the database can reduce the measured sensitivity by 15%-25%. Because of the strong dependence of measured performance on the testing database, it is difficult to estimate reliably the accuracy of a CAD scheme. Furthermore, it is questionable to compare different CAD schemes when different cases are used for testing. Sharing databases, creating a common database, or using a quantitative measure to characterize databases are possible solutions to this problem. However, none of these solutions exists or is practiced at present. Therefore, as a short-term solution, it is recommended that the method used for selecting cases, and histograms or mean and standard deviations of relevant image features be reported whenever performance data are presented.

Comparison of bilateral-subtraction and single-image processing techniques in the computerized detection of mammographic masses

RATIONALE AND OBJECTIVES

Identification of regions as possible masses on digitized screen film mammograms is an important initial step in the computerized detection of breast carcinomas. Possible masses may be initially extracted using criteria based on optical densities, geometric patterns, and asymmetries between corresponding locations in right and left mammograms. In this study, the usefulness of information arising from mammographic asymmetries for the identification of mass lesions is investigated.

METHODS

Two techniques are investigated--a nonlinear bilateral-subtraction technique based on image pairs and a local gray-level thresholding technique based on single images. Detection performances obtained with the two techniques in combination with various feature-analysis techniques are evaluated using 154 pairs of mammograms and compared using free-response receiver operating characteristic (FROC) analysis.

RESULTS

The nonlinear bilateral-subtraction technique performed better than the local gray-level thresholding technique.

CONCLUSION

The incorporation of asymmetric information appears to be useful for computerized identification of possible masses on mammograms.

Evaluation of imaging properties of a laser film digitizer

In this paper we provide a quantitative assessment of the basic imaging properties of a laser film digitizer. The characteristic curve of the digitizer was determined in terms of the relationship between input optical density and output pixel value. Spatial resolution of the laser digitizer was characterized using the presampling modulation transfer function (MTF), which was measured using a curve fitting technique with an angulated slit. For the noise analysis, we compared the Wiener spectra of uniformly exposed film samples before and after digitization. The effects of different sampling distances and scanning directions were investigated. Our results show that the characteristic curve of the laser digitizer was linear. The presampling MTFs of the digitizer were similar at different sampling distances and were substantially greater in the vertical scanning direction than in the horizontal direction. The noise of the digitized film sample was mainly affected by the presampling MTF and structure noise of the digitizer.

Computerized detection of masses in digital mammograms: analysis of bilateral subtraction images

A computerized scheme is being developed for the detection of masses in digital mammograms. Based on the deviation from the normal architectural symmetry of the right and left breasts, a bilateral subtraction technique is used to enhance the conspicuity of possible masses. The scheme employs two pairs of conventional screen-film mammograms (the right and left mediolateral oblique views and craniocaudal views), which are digitized by a TV camera/Gould digitizer. The right and left breast images in each pair are aligned manually during digitization. A nonlinear bilateral subtraction technique that involves linking multiple subtracted images has been investigated and compared to a simple linear subtraction method. Various feature-extraction techniques are used to reduce false-positive detections resulting from the bilateral subtraction. The scheme has been evaluated using 46 pairs of clinical mammograms and was found to yield a 95% true-positive rate at an average of three false-positive detections per image. This preliminary study indicates that the scheme is potentially useful as an aid to radiologists in the interpretation of screening mammograms.

Comparison of imaging properties of a computed radiography system and screen-film systems

To compare the diagnostic quality of images obtained with a computed radiography (CR) system based on storage phosphor technology with that obtained with conventional screen-film systems, a dual-image recording technique was devised. With this technique, a CR imaging plate is placed behind a screen-film system in a conventional cassette. This makes it possible to obtain two images simultaneously, one from each system, in a clinical examination with the same patient positioning, the same degree of patient motion, the same geometric unsharpness, and no additional exposure. The modulation transfer functions (MTFs) of the CR system with and without the dual-image recording technique were greater at low frequencies, but lower at high frequencies, that the MTFs of the screen-film systems used. The noise Wiener spectra of the CR images at the plane of the imaging plate were greater than those of the screen-film systems, but were comparable to those of the screen-film systems at the plane of the printed film due to the reduction in image size. Clinical chest images obtained with the dual-image recording technique appeared comparable, probably because of the image size reduction and the use of mild unsharp mask processing.

Direct comparison of conventional and computed radiography with a dual-image recording technique

To compare the image quality of computed radiographic (CR) with conventional screen-film images, the authors used a dual-image recording technique. Images were simultaneously acquired with a conventional screen-film combination and a storage-phosphor imaging plate loaded into a single cassette. Wiener spectra and modulation transfer function were compared for both image types. A preliminary observer performance test was conducted with chest images obtained with the dual-image recording technique on portable and fixed equipment. Analysis of physical parameters and observer test results suggests that the conventional screen-film system can provide slightly superior image quality, although the CR system has some advantage in bedside applications.

Data compression: effect on diagnostic accuracy in digital chest radiography

High-resolution digital images make up very large data sets that are relatively slow to transmit and expensive to store. Data compression techniques are being developed to address this problem, but significant image deterioration can occur at high compression ratios. In this study, the authors evaluated a form of adaptive block cosine transform coding, a new compression technique that allows considerable compression of digital radiographs with minimal degradation of image quality. To determine the effect of data compression on diagnostic accuracy, observer tests were performed with 60 digitized chest radiographs (2,048 x 2,048 matrix, 1,024 shades of gray) containing subtle examples of pneumothorax, interstitial infiltrate, nodules, and bone lesions. Radiographs with no compression, with 25:1 compression, and with 50:1 compression ratios were presented in randomized order to 12 radiologists. The results suggest that, with this compression scheme, compression ratios as high as 25:1 may be acceptable for primary diagnosis in chest radiology.

Measurement of the presampling modulation transfer function of film digitizers using a curve fitting technique

A curve fitting technique combined with an angulated slit image has been developed for the measurement of the presampling modulation transfer function (MTF) of film digitizers. The noisy line spread functions (LSFs) acquired from an angulated slit image are fitted using a combination of two functions by means of a nonlinear least-square fitting technique. The parameters in the model function for each LSF are obtained by minimizing the residual root mean square (RMS), and then averaged over all the LSF fittings. The resulting analytical function is representative of the continuous presampled LSF. We have found that a combination of Gaussian and exponential functions provides a good fit to the LSFs obtained with film digitizers. The corresponding analytical Fourier transformation of the model function yields the presampling MTF, without Nyquist frequency limitation. Measurements of spatial resolution properties using this method were performed for two laser scanners and an optical drum scanner.

Pulmonary nodules: computer-aided detection in digital chest images

Currently, radiologists fail to detect pulmonary nodules in up to 30% of cases with actually positive findings. Diagnoses may be missed due to camouflaging effects of anatomic background, subjective and varying decision criteria, or distractions in clinical situations. We developed a computerized method to detect locations of lung nodules in digital chest images. The method is based on a difference-image approach and feature-extraction techniques, including growth, slope, and profile tests. Computer results were used to alert 12 radiologists to possible nodule locations in 60 clinical cases. Preliminary results suggest that computer aid can improve the detection performance of radiologists.