The ray density estimation of a CT system by a supervised learning algorithm

Since the development of the CT scanner, the faster scan time was the highest priority to achieve, and therefore, there have been a lot of efforts to reduce the scan time. The first generation CT scanner (Figure 1(a)) uses one pencil beam, and the projection data were acquired by translating the x-ray source and detector linearly. After the completion of the linear measurements, the x-ray source and detector rotated to the next angular position to acquire the next set of measurements. A faster scan time could be achieved by using multiple pencil beams. Since this second generation CT scanner (Figure 1 (b)) used a small fan beam, the data acquisition time was reduced by the same factor of the increased detector numbers. However, a translation-rotation principle was still employed for data acquisition. The third generation scanner (Figure 1 (c)) used a large number of detectors and a single source. Since the detector size was sufficient to cover the entire object, a translation-rotation principle was not employed. The elimination of the translation step reduced the scan time significantly, and nearly all of the state-of-the-art scanners on the market today are third generation. The potential problem of the third generation system is the high detector cost, and high scatter-to-primary ratio. In order to combat these problems, a different type of CT geometry can be imagined as shown in Figure 1 (d). In this system, instead of using one source and many detector cells, it uses many sources and smaller number of detector cells. For example, if the third generation CT system uses 1000 detector cells, the new system (Figure 1(d)) may employ 50 detector cells, and 20 sources so that the detector cost can be reduced by a factor of 20.