Evaluation of a high-sensitivity dual-head small animal PET scanner in lesion detection by using computer observer

We have developed a prototype small-animal PET scanner that employs two large-area planar HRRT detector heads arranged in a compact configuration [1], [2]. This prototype has a measured central sensitivity of ∼30%. For rodent imaging at the typical ranges of radioactivity, the measured noise-equivalent count rates of the prototype are also substantially higher than those reported in the literature for other small-animal PET systems. Due to its compact configuration, the prototype exhibits substantial depth-of-interaction (DOI) blurring. We have successfully modeled the DOI blurring of the prototype and used the resulting model to correct for the DOI blurring in reconstruction. After correction, we showed that the prototype can yield an image resolution of about 1.2mm. Since correcting for resolution blur in reconstruction is known to increase the image noise, the question regarding the prototype’s “net” sensitivity — which includes both the detection sensitivity of the scanner and the noise-amplification effects in reconstruction — arises. To address this question is a challenging task. In our opinion, it requires an objective means for evaluating the performance characteristics of an image device in terms of the quality of the resulting images. To the best of our knowledge, a formal framework of this kind has yet to the developed and agreed upon. In this work, we approach this important question by evaluating the equivalent sensitivity of our prototype with respect to a theoretical reference system and a numerical observer in terms of a certain lesion-detection task. Our results indicate that for detecting 1mm lesion at center of a rat-sized cylinder, our prototype has an equivalent sensitivity, with respect to the reference system and the numerical observer, of ∼18%. This equivalent sensitivity increases to reach ∼22% as the lesion size increases.