George Revesz

Psychophysical Studies of Detection Errors in Chest Radiology

In 62 of 124 cases analyzed, there occurred a failure to detect a pulmonary nodule which was retrospectively noted only after detection at a subsequent examination. The characteristics of the lesions and their surrounds were measured, and a computer analysis used, to identify that combination of parameters (the conspicuity, K) which was best capable of separating the group into populations of missed and detected lesions.

The effect of verification on the assessment of imaging techniques.

Revesz G, Kundel HL, Bonilatibus M. The effect of verification on the assessment of imaging techniques. Invest Radiol 1983; 18: 194-198. In order to measure the accuracy of a diagnostic imaging technique, the correctness of each observer decision must be determined. This is often difficult in clinical cases, and various strategies are frequently used to approximate the truth. Experiments using chest radiographs as an example are reported to show the pitfalls in those strategies. To assess relative accuracies of three different radiographic techniques, chest radiographs were taken of 66 patients with each of the three techniques, and the films were evaluated by six radiologists. Their findings were then scored by comparing them with the correct decision defined by each of the following methods: majority vote, consensus opinion, expert judgment, feedback review, and clinical/pathologic proof. The findings snowed that any one of the techniques could be shown to be better than the other, depending on how the truth was defined. It is concluded, therefore, that strategies that define the presence or absence of disease only by the diagnostic tests under evaluation are inadequate. The truth must be determined by clinical or pathologic proof and follow-up data.

Contrast gradient and the detection of lung nodules.

The effects of edge gradient and blur on the detection pulmonary nodules were studied, using a test series of 175 artificial nodules on 30 chest films. The nodules were of homogeneous texture, about 1 cm in diameter and had a mean contrast of .12. Six observers viewed the film set, scoring both location and confidence. The sensitivity or the probability of a true-positive response was linearly related to the blur and a modified ROC analysis resulted in a family of ROC curves with the index of detectability monotonically decreasing with increased blur.

Femoral neck torsion angle measurement by computed tomography.

To establish the accuracy and reproducibility of measuring femoral neck torsion angles (FNTA) on CT images, this angle was measured on 12 dry femora directly on the gross specimens, on plain radiographs, and on CT images. The selection of the CT level, both at the femoral neck and the condyles, that may affect the measurement of FNTA was analyzed. The CT measurements of FNTA were found to have a reproducibility of 2.5 and 0° mean offset error when compared with readings on plain radiographs. A new CT method was developed to measure this angle particularly in patients with coxa valga in whom this measurement is difficult if not impossible to accomplish by current methods. For this purpose composite CT images were made of the femoral head and the femoral shaft, immediately below the lesser trochanter. On these images the line connecting the centers of the femoral head and shaft were inferred to represent the cervical plane. In comparison with the standard CT method, this new method had the same reproducibility, although it slightly overestimated this angle by 2°. This degree of error is negligible when compared with the normal variations of this angle, which ranges from −20 to + 38°. Furthermore, it compares favorably with the limited accuracy of derotational osteotomy surgery.

Thermography and ultrasound in detection and diagnosis of breast cancer

Although thermography has not proven to be of significant value in detecting early breast cancer as used in the National Breast Cancer Detection Program, it can represent a high risk indicator and can be useful in combination with other factors in reducing the number of women who should receive routine mammograms for screening purposes. In breast cancer patients, thermography has been shown to have prognostic value. “Cold” breast cancers have an increased survival as compared to those with increased infrared emission in some series. The use of diagnostic ultrasound to date has been generally limited to evaluation of breast cancer or masses found by either palpation or mammography. Recent improvements in technology with the use of multiple transducers, higher frequencies, through‐transmission techniques, and computerization have resulted in finding lesions greater than one centimeter in diameter. Lesions below this size and small calcifications have not so far been generally demonstrated successfully because of the limitation of ultrasound resolution.

The effect on medical metal implants by magnetic fields of magnetic resonance imaging

Forces and torques, due to a 0.3 Tesla magnetic field were evaluated on ten hip prostheses and ten hemostat clips. Measurements were performed with an instrument utilizing the movement of a laser beam caused by the deflection of a canti-lever. The results indicate effects, if any, to be smaller than the instruments sensitivity which, at its highest, was 7 mg of force and 125 mg·cm of torque.

The effect of magnetic resonance imaging on metal spine implants.

In Part 1 of this study, the forces and torques exerted on metallic spine implants by a 0.3-T magnetic field were evaluated using a cantilever system sensitive to forces as low as 250 mg and torques as low as 4 g/cm. Results Indicated that the 0.3-T magnification field did not cause forces or torques that could be measured within this range. It is concluded that patients with spine implants may safely undergo magnetic resonance scanning. In Part 2, the magnetic resonance scans from 33 patients (61 studies) then were reviewed to determine which, if any, parameters could be adjusted to minimize artifact. Magnetic power, slice thickness, plane of scan, and pulse sequence were considered. Plane of scan was found to be the most significant parameter in achieving useful studies, with the sagittal plane being the preferred orientation. Scans of patients who had posterior wiring were more likely to be useful, while scans of patients with spinal rods usually showed a large amount of artifact.

Patient flow and utilization of resources in a diagnostic radiology department.

Operation of the radiology department in a large university hospital has been analyzed in terms of patient movement and the utilization of technicians and examination rooms. Analysis of patient movement involved collating and statistically evaluating data on the time patients spent at each stage of the process, while utilization required the accounting for technician activities and examination room occupancies during all periods of the day. Comparison of the results with data from other similar institutions shows good agreement and indicates the possibility of increasing the utilization of available resources while decreasing the time patients spend in the department.

Evaluation of a television image processing system.

An image processing device using television and a scanning contrast enhancer has been evaluated by study of error rates of 13 observers viewing a standard series of chest x-rays. A technique for creating artificial lesions which simulate parenchymal masses was developed, allowing known densities to be put in known locations in otherwise normal chest x-rays. Detection was chosen as the criterion of performance. An image processor can be considered useful if percentage error is decreased. Observers were shown a series of 54 chest x-rays on a conventional viewbox and on the television monitors of the image processor. A statistical comparison of error rates was used to evaluate the system. The results show that the unprocessed television display degraded performance considerably. The addition of contrast enhancement improved error rates but direct viewing was still superior to image processing. A detailed study of error distribution showed that, using image processing, observers made considerably poorer judgments on lesions in the lower chest areas than in the upper. In the upper areas there was a trend toward lower error rates with image processing than with direct viewing although a statistically significant difference could not be demonstrated. The possible reasons for the error rate variations are discussed.

Spatial Frequency Spectra of Radiographic Images

Abstract An analysis of the input to image processing systems is presented for input images recorded on radiographic film. The effects of combining simple structural elements to produce complex images are discussed. Mathematical models for idealized radiographic images of ribs, pulmonary nodules, and blood vessels are formulated and the effect of various combinations of these models on the frequency spectra of the complex image is examined in detail. The validity of these theoretical predictions is demonstrated by using a radiograph of real structures. The implication of these results for image processing by matched filters is discussed.