The aim of this work was to introduce and validate a fully automated approach for liver recognition in whole-body 18FDG Positron Emission Tomography (PET) scans, so as to enable the reproducible calculation of tracer uptake in it
The algorithm starts by co-registering the PET scans in the geometrical centre of mass and rotates them according to the patient’s position. 3D SUV matrices are then projected in the coronal direction and convoluted with a Heaviside function. The largest area of higher homogeneity is then identified in the projected images. Two additional cuts on the maximum distance between the geometric center of mass of the patient to the liver and on the right part of the body are also applied. The algorithm then extracts the coordinates of the liver position in the anterior-posterior direction and calculates the maximum, average and standard deviation SUV values in a 5 cm radius sphere centered in this position. The SUV in the liver was also measured manually by drawing a VOI in a central portion of the liver at least 4 cm far from the liver dome and 2 cm far from the liver’s edge
In the 630 analyzed PET scan, coming from more than 40 PET centers distributed in 21 countries, the algorithm identified the liver in 97,3% of the cases. In 1,4% of the cases the algorithm mismatched the liver with the right lung and in 1,3% with other structures. The liver SUV evaluated by the algorithm and manually measured were not statistically different, (1.90±0.52) and (2.01±0.57)
The algorithm demonstrated to be robust and efficient in identifying the liver and its average SUV values were compatible with the manual measurement. The algorithm could be useful in evaluating automatically image quality in a standard setting, to retrieve average liver values for lesion to liver ratio or as initial seed in segmentation algorithm. Further work is in progress to identify other physiologically relevant organs or tissue

Scarica PDF