F V Gabor1, F L Datz, P E Christian. 1. Department of Radiology, University of Utah School of Medicine, Salt Lake City 84132.
Abstract
UNLABELLED: An expert system was developed that interprets ventilation-perfusion lung scans. The use of such scans for suspected pulmonary embolism is ideal for computer-assisted diagnosis by expert systems. The data are digital, only a single disease entity is diagnosed or excluded, and well-established diagnostic criteria already exist for visual interpretation that can be easily integrated into an expert system. METHODS: This expert system is divided into two modules. The first module is responsible for image analysis. Analysis was performed on the eight standard perfusion images and on single-breath, equilibrium and 3-min washout ventilation images. Each image was analyzed for the presence of regional perfusion or ventilation defects, as determined by pixel values that fell 2.2 s.d. below the mean (or above the mean in the case of washout images) compared with a database of normal studies. The defect size, segment involved and number of defects were determined. Ventilation and perfusion images were then compared to determine whether defects were matched or mismatched. The second program module applied the modified Biello's criteria to the data and categorized the scan as normal to low, intermediate or high probability. RESULTS: A total of 80 patients were prospectively studied. An 81% (65 of 80) correlation was obtained when the results of the expert system were compared with visual interpretations made by three experienced nuclear medicine physicians. CONCLUSION: This study shows that the interpretation of ventilation-perfusion lung scans by an expert system is possible. The technique holds the promise of reducing interobserver variability and assisting less experienced observers in the interpretation of such scans.
UNLABELLED: An expert system was developed that interprets ventilation-perfusion lung scans. The use of such scans for suspected pulmonary embolism is ideal for computer-assisted diagnosis by expert systems. The data are digital, only a single disease entity is diagnosed or excluded, and well-established diagnostic criteria already exist for visual interpretation that can be easily integrated into an expert system. METHODS: This expert system is divided into two modules. The first module is responsible for image analysis. Analysis was performed on the eight standard perfusion images and on single-breath, equilibrium and 3-min washout ventilation images. Each image was analyzed for the presence of regional perfusion or ventilation defects, as determined by pixel values that fell 2.2 s.d. below the mean (or above the mean in the case of washout images) compared with a database of normal studies. The defect size, segment involved and number of defects were determined. Ventilation and perfusion images were then compared to determine whether defects were matched or mismatched. The second program module applied the modified Biello's criteria to the data and categorized the scan as normal to low, intermediate or high probability. RESULTS: A total of 80 patients were prospectively studied. An 81% (65 of 80) correlation was obtained when the results of the expert system were compared with visual interpretations made by three experienced nuclear medicine physicians. CONCLUSION: This study shows that the interpretation of ventilation-perfusion lung scans by an expert system is possible. The technique holds the promise of reducing interobserver variability and assisting less experienced observers in the interpretation of such scans.