Evaluation of bi-level image converting methods for nuclear medicine image database stored in the IS&C magneto-optical disk.

1. INTRODUCTION. We have developed a report and imaging management system for nuclear medicine. The report and image data are stored in the IS&C magneto-optical disk (IS&C MOD). Nuclear medicine image data are relatively small, but they are large enough to create problems when being transported over a low-speed Hospital Information System (HIS) network. Moreover, gray scale image output devices (i.e., high-resolution displays, sonoprinters) are expensive. If high quality bi-level (black and white) images were available, images could be transferred through low-speed network with inexpensive bi-level terminals or conventional page-printers. We examined images using several bi-level image conversion techniques [1, 2] in order to determine how useful the bi-level images are and to assess their suitability for use in nuclear medicine. The images were compared with original film images by ROC analysis. The modified minimized average error method was found to be superior to other methods in bone and gallium images. Its A-values are 0.83 in gallium scan and 0.85 in bone scan. 2. MATERIALS AND METHODS. Our system consists of three digital gamma cameras, a nuclear medicine data processing unit, two UNIX stations, and two personal computers. The computers and the data processing unit are connected via Ethernet and each computer has an IS&C MOD unit. Image and report data are stored in the IS&C MOD. The computers are also connected off-line through the IS&C MOD. To evaluate image quality, bone scan images and gallium scan images (1024x512, 2048x1024 pixels and 16 bits depth) were converted to bi-level images (same pixels and one bit depth) using four methods: dither method, minimized average error method (MAE)[1], modified regional adaptation method, and constrained average method. For the dither methods, three sets of dither matrix (Fatting's, Bayer's and our original matrix) were employed. The computer images were compared with original film images by ROC analysis. 3. RESULTS. The converted bi-level images were 1/16 the size of the originals. They were transferred via Ethernet, displayed on the monochrome display, and printed using a conventional page-printer (240 dpi or 300 dpi). The A-values of the original gallium and bone images on the films were 0.89 and 0.94. Modified MAE method gave better results than other methods tested, with the gallium image using this technique reading 0.83 on the A-scale; and the bone images were 0.85. 4. DISCUSSION. Generally, nuclear medicine images require lower spatial and gray level resolution than other computer imaging techniques. We compared four bi-level image conversion methods in order to know which methods are suitable to nuclear medicine images. The ROC analysis was employed to evaluate these image qualities. The MAE methods made particular texture pattern as artifacts in the intermediate gray level area. But it can express homogeneity the neutral level. The dither method images are coarse texture. It affects detectabilities of subtle abnormal findings. 5. CONCLUSIONS. The A-z values of the modified MAE images were comparable to those of original film images. Although bone images give slightly lower values, many abnormal findings can be ascertained on bi-level images. High quality bi-level image techniques are considered to be useful for nuclear medicine image database systems.