Abnormality manipulation of PET images

Objectives Image perception studies using clinical PET images can be challenging because of the difficulty in compiling control and target cases. We have developed and tested software that allows the removal and insertion of abnormalities so that perception case sets can be rapidly constructed.

Methods The methods for manipulating PET images have been developed from similar approaches used on CT data. A review module allows the logging and capture of abnormalities to a library for insertion into selected regions of other PET studies. Abnormal areas are removed by locating elliptical masks over the abnormality and an appropriate replacement region in the surrounding tissue. Alternative forced choice experiments were performed with 7 experienced PET physicians to evaluate the performance of the software. 27 cases of 5 axial slices had abnormalities removed and were displayed with 3 unaltered data sets. The task required the observer to select the altered images from among the 4 displays. Each case was used twice with order scrambling for a total of 54 cases.

Results The mapping software is complete and allows the easy review and logging of PET CT data. For the PET abnormality removal experiment, observers on average correctly called the altered cases in only 19.4 out of 54 (chance: 13.5) and correctly called repeats in only 4.7 out of 27 cases (chance: 1.7). Further, the consensus observer (4 or more observers with the same selection) was consistent with chance selection being correct only 13 times for all cases and correct for only 1 of the repeats.

Conclusions The set of tools developed for reviewing and manipulating PET images are useful for creating case sets for PET image perception studies using real clinical images. The removal software has been successfully tested with human observers and observer experiments with inserted abnormalities will be completed in the near future.

Research Support This work is supported by the NIBIB (5R01EB006638-03)