RAI Labs Duke
Breast Tomo Clinical Trial | Chest Stereo/BCE Clinical Trial | Chest Tomo Clinical Trial
Quant. Image | Emerg. Quant. Imaging | Perf. Metrology | Clinical Trials | Emerg. Clinical


Clinical trial of chest stereo and biplane correlation imaging

The early detection of lung cancer has been one of the outstanding challenges in radiographic imaging, the significance of which can be discerned only by considering the fact that lung cancer remains the leading cause of cancer death in the US, surpassing breast, prostate, colon, and cervical cancers combined. Prior research has shown that interference of the anatomical structure is the dominant factor in the low detection of early lung cancer in radiographic images. Computer aided detection (CAD) is an advanced image processing technique that can somewhat remedy that limitation by enabling a full evaluation of an image for the presence of cancer. However, the fundamental limitations of anatomical noise still persist.


Original chest radiograph containing lung nodules.
Green circles indicate actual lesions, pink areas are computer-detected suspicious regions.

A new imaging paradigm, correlation imaging (CI), is pursued at RAI Labs for improving the detection of subtle lung nodules. In CI, two or more digital images of the thorax are acquired within a short time interval from two slightly different posterior projections. The image data are then incorporated into an enhanced CAD algorithm in which nodules are detected by examining the geometrical correlation of the detected signals in multiple views. Angular information is used to minimize the limiting influence of anatomical noise by identifying and positively reinforcing the nodule signals, which remain relatively constant against a variation in the background structure. This approach does not promise to completely eliminate anatomical noise (as CT does), but aims to cost-effectively and dose-effectively reduce its influence with little or no increase in the patient dose. Using correlation of signals between multiple views to identify “true” signals, CAD is used at high sensitivity levels, lowering the detection thresholds, without an undesirable increase in the number of false positives. This hybrid approach of utilizing angular information in conjunction with digital acquisition and CAD addresses all three major obstacles to the detection of subtle lung nodules; the angular information reduces the effects of anatomical noise, the high signal-to-noise ratio of digital acquisition assures sufficient nodule contrast, and CAD incorporates a complete search. The figure above demonstrates a schematic of an acquisition system. The images show the results of CI on a chest phantom. The highlighted areas are suspect lesions identified by CI, while circles show the location of true lesions.


Fig. 3: The completed prototype BCI image acquisition system.

We have designed, developed, and perfected a new prototype system that is able to acquire multi-projection x-ray images from any direction (vertical, horizontal, or arbitrary combinations of the two) within a ± 20 degree angular span (Fig. 3). The tube movement is provided by our mechanical apparatus empowered by 3 actuators. The tube moving apparatus is fully synchronized with the detector and the x-ray generator. We have also validated the mechanical reliability and reproducibility of the system.

With a validated prototype imaging system, we have been making steady progress towards acquisition of human subject data. As of now, we have captured BCI images from 86 human subjects. Each subject was imaged at three projections of -3, 0, and +3 degree horizontal angulations. This enables stereoscopic viewing of the image data. The total acquisition time was about 8 seconds during which the subject was asked to hold still and take a deep breath. Fig. 4 shows some of the images of the human subjects.

We have streamlined our stereoscopic presentation of bi-plane images, enabled by a collaboration with Planar display systems,. The system is based on two grayscale, 5 mega pixel LCD displays separated by a semi-transparent mirror enabling the projection of a stereo pair toward the observer (Fig. 5). The observer would then be able to separate the two images using a passive polarizing glass with minimal impact on perceived brightness. With three images acquired from each human subject at -3, 0, and +3 degree horizontal angulations, three image pairs can be formed for stereo visualization. We have further finalized a graphical user interface that enables easy navigation of radiologists through the stereo image datasets.

We have recently conducted the first observer study of BCI/Stereo imaging system using our first 72 human subject cases. Four experienced chest radiologists read all cases monoscopically and stereoscopically in two separate reading sessions. All readings took place with the same display hardware. The sessions were one month apart to minimize the likelihood of memory effects. Each observer assigned a score between 0 to 100 to each case indicating his/her confidence that a lung nodule is present. The data were processed using a Multi-Reader-Multi-Choice (MRMC) methodology to assess the ROC curves and the area under the ROC. Some results are illustrated Fig. 6. Overall, the results indicate that for each observer, a higher diagnostic accuracy can be achieved with the stereoscopy/BCI technology. These results are without CAD input, which is expected to provide a higher differential improvement. These findings are consistent with prior work in mammographic displays.




Fig. 4: Example clinical images of three healthy volunteers acquired at +3° (a), 0° (b), and -3° (c) about the PA orientation using a recently developed multi-projection chest imaging system.

Fig. 5: The prototype high-resolution grayscale stereo display system used in the observer study.
Fig. 6: ROC results (plotting sensitivity as a function of 1-specificity) reflecting the averages from four experienced radiologists.


Breast Tomo Clinical Trial | Chest Stereo/BCE Clinical Trial | Chest Tomo Clinical Trial
Radiographic Tech. | Low-Cost Tomo | Optimiz. of Rad. Ther. | Personal CAD | Adapt. Educ.