ABSTRACT
Motion artifacts have been identified as a problem in medical tomography systems. While computed tomography (CT) imaging has been getting faster, there remains a need to detect and compensate for motions in clinical follow-up of neurological patients (multiple sclerosis, tumors, stroke, etc.), in cardiac imaging, and in any area in which failing to detect a motion artifact may lead to misdiagnosis. We have developed a novel algorithm to detect motion in brain images. The algorithm deals with detecting and isolating motion in the object domain using only the information available in the sinogram domain. The new "opposite ray algorithm" (ORA) addresses the issue of motion in the interior elements of the object. The ORA combines information from projections that are opposite in space and separated in time to isolate and identify the motion. A sinogram of motion is created, integrated and reconstructed to isolate the moving component. The algorithm can be used with conventional clinical scanners employing quarter-detector offset. The significant effect of quarter-detector offset on the ORA is investigated. The effects that a finite beamwidth and noise have on the ORA are also investigated. Both the similarity index and a correlation coefficient are used to evaluate the algorithm. The algorithm is successful when applied to cases exhibiting translational and translational-rotational motion. A similarity index of 0.88 is obtained in a typical case with both translational and rotational motion. Further development is recommended in the deformation case.
Subject(s)
Movement/physiology , Tomography, X-Ray Computed/methods , Algorithms , Artifacts , Brain/diagnostic imaging , Brain/physiology , Equipment Design , Fourier Analysis , Heart/diagnostic imaging , Heart/physiology , Humans , Image Processing, Computer-Assisted , Lung/diagnostic imaging , Lung/physiology , Models, Theoretical , Phantoms, Imaging , RotationABSTRACT
We present a new method of registering three dimensional image volumes of the brain of a given patient acquired at different times or with different imaging modalities, or both. Registration is an essential requirement for fusing the data from the two image sets so as to either increase the available information by exploiting complementary imaging modalities, or to measure small changes over time for prognostication, disease assessment, etc. The new technique exploits an external, removable, remountable reference frame which is attached to the head. Computer vision techniques are used to determine the positions of fiducial marks in every image. The transformation required to map each image of one image volume onto the other image volume is developed using the theory of quaternions. The results indicate that the new technique is robust and practical in a clinical setting.
Subject(s)
Algorithms , Brain/anatomy & histology , Brain/diagnostic imaging , Image Processing, Computer-Assisted/methods , Magnetic Resonance Imaging , Tomography, Emission-Computed, Single-Photon , Humans , Image Processing, Computer-Assisted/instrumentation , Models, Theoretical , Posture , Reproducibility of Results , Rotation , Stereotaxic TechniquesABSTRACT
Landmarks, or certain characteristic reference points, on cephalograms are used as a diagnostic aid employed in treatment planning by orthodontists. This work presents an algorithm for recognizing some anatomical features and locating landmarks on lateral skull X rays (cephalograms) using digital image processing and feature recognition techniques. A cephalogram is digitized and stored in a computer memory. Prefiltering is applied to remove image noise. The bony and flesh profiles of jaw and front face are traced. Using these profiles the algorithm locates 17 points on the image, some on bony features and others on soft tissue. To locate these points, edge-enhancement, thresholding, and edge-detection techniques are applied. The algorithm can be run on an IBM compatible microcomputer.
Subject(s)
Cephalometry/methods , Image Processing, Computer-Assisted , Microcomputers , Orthodontics/methods , Algorithms , Image Enhancement/methods , Research DesignABSTRACT
The purpose of this study was to compare the diagnostic performance of a digital radiography system that uses 6- and 8-bit displays with conventional D-speed film for the detection of simulated periodontal bone lesions. Eleven human hemimandibles were used as specimens. Simulated lesions were created at the buccal cortical plate in the marginal bone area with the use of a round bur 1.4 mm in diameter. Lesions were created in a defined sequence to preclude visual cues as to the depth of the lesions. Lesion size progressed in 0.5 mm increments. At each stage the mandibles were imaged with a Sens-A-Ray system (REGAM Medical Systems AB, Sundsvall, Sweden) and D-speed film. Exposure parameters for each specimen/receptor combination were standardized by either the mean optical density or mean gray value at the approximal crestal bone area. Film images and digital images displayed with 64 and 256 gray levels were presented to six observers for evaluation. Observers were ask to rate their confidence as to the presence or absence of a lesion using a 5-point confidence scale. A total of 96 lesion sites and 96 control sites were presented to the observers. Receiver operating characteristic curves were generated for each system. The area under the curve was used as the index of diagnostic accuracy. The mean receiver operating characteristic areas for 6-bit and 8-bit displays and D-speed film were 0.746 +/- 0.043, 0.717 +/- 0.056 and 0.742 +/- 0.059, respectively. Analysis of variance was used to compare the means. No statistical difference was found between any of the three image displays (p > 0.05).(ABSTRACT TRUNCATED AT 250 WORDS)
Subject(s)
Periodontal Diseases/diagnostic imaging , Radiographic Image Enhancement/methods , Analysis of Variance , Contrast Sensitivity , Dental Caries/diagnostic imaging , Humans , Mandible/diagnostic imaging , ROC Curve , Reproducibility of ResultsABSTRACT
The degree of venous beading in ocular fundus images has been shown to be a more powerful predictor of conversion to proliferative diabetic retinopathy than any other type of retinal abnormality. Further, the degree of venous beading has been shown to be well correlated with disease progression. An algorithm for automated grading of venous beading in digitized ocular fundus images is described. Thresholding is used to extract a rough silhouette of the vein. Morphological closing is used to fill any holes in the silhouette arising from either the central light reflex or noise. The silhouette is then "thinned" to find vein centerlines. Each centerline is partitioned into fixed-length segments of 32 pixels. Vein diameters are measured as a function of distance along each segment with the aid of the local centerline orientations. The resulting diameter data are then interpolated and resampled to generate diameter data at constant sampling intervals. A fast Fourier transform is performed on the resulting data to determine the magnitude spectrum of vein segment diameter. A venous beading index is calculated from the distribution of vein diameter frequency components. Performance of the new algorithm is compared to the currently accepted clinical practice of manual grading in a pilot clinical study of 51 subjects. The algorithm is seen to perform well.
Subject(s)
Algorithms , Diabetic Retinopathy/diagnostic imaging , Radiographic Image Enhancement , Retinal Vein/diagnostic imaging , Diabetic Retinopathy/pathology , Fourier Analysis , Fundus Oculi , Humans , Phlebography/methods , Retinal Vein/pathologyABSTRACT
It is known that contracting muscle makes low frequency sound vibrations. Small vibrations of uncertain origin are found over resting muscle. These could be shown to be of muscle origin if they significantly diminish in response to agents expected to decrease muscle activity. Thiopental, propofol, and neuromuscular-junction blocking muscle relaxants have such properties. Twenty-one subjects slated for elective surgery for which they would routinely be anesthetized and paralysed gave informed consent to having a small accelerometer taped upon their supine biceps (9 subjects), or volar forearm (12 subjects). Recordings were made in four stages while subjects: (i) lifted a 2-kg weight just off the sponge armrest on which their outstretched arm lay; (ii) relaxed their arm in the awake state prior to anesthesia; (iii) had anesthesia induced with intravenous thiopental (n = 11) or propofol (n = 10); and (iv) were paralysed. Recordings were digitised at 172-Hz and 6-s segments fast Fourier transformed (FFT). Total signal power, as determined by the area under the power spectrum, was significantly different (p < 0.05) in all stages for the biceps and in all but stages (iii) from (iv) in the forearm. It appears that resting muscle generates measurable vibrations.
Subject(s)
Anesthesia, General , Muscle, Skeletal/physiology , Neuromuscular Blocking Agents/pharmacology , Adolescent , Adult , Female , Forearm , Humans , Male , Middle Aged , Muscle, Skeletal/drug effects , Sound , VibrationABSTRACT
OBJECTIVE: To describe a cardiac output measurement using a new method to derive and analyze the long-axis ballistocardiogram that is less invasive than pulmonary artery thermodilution. DESIGN: Prospective physiologic study. SETTING: Intensive care unit of The Halifax Infirmary, a teaching hospital of Dalhousie University, Halifax, NS. PATIENTS: Thirty-nine patients in sinus rhythm with pulmonary artery thermodilution catheters or radial artery catheters in place. The first 30 subjects were the "learning set" and the next 9 were the "test set." INTERVENTIONS: A small (54-g) accelerometer was taped on the patient's chest. OUTCOME MEASURES: Measurements of time and amplitude coordinates of the acceleration and radial artery pressure wavepeaks, as well as anthropometric information. RESULTS: A stroke volume prediction equation was generated (R2 = 0.76) from the learning set. This equation was applied to the test set and correlated with the pulmonary artery thermodilution-derived stroke volumes (R = 0.79). Stroke volumes were compared using a previously described statistical method: a) bias (predicted > thermodilution) = 0.03 mL (95% confidence interval [CI] -4.2 to 4.8 mL); b) lower limit of agreement = -21 mL (95% CI -29 to -13 mL); c) upper limit of agreement = 22 mL (95% CI 14 to 29 mL). Of derived stroke volumes, 82% were within 15 mL of pulmonary artery thermodilution-derived values. CONCLUSIONS: The sternal acceleration ballistocardiogram combined with hemodynamic and demographic data in a probabilistic model shows promise of providing a less invasive measure of cardiac output than thermodilution.