This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Quality assurance, Screening, Computer Applications-Detection, diagnosis, CAD, CT, Lung
Authors:
J. C. Marasinghe1, M. Ohkubo1, A. yoshida1, A. Kayugawa1, K. Murao2, T. Matsumoto3, S. Niizuma1, S. Sone 4, S. Wada1; 1Niigata/JP, 2Tokyo/JP, 3Chiba/JP, 4Nagano/JP
DOI:
10.1594/ecr2015/C-1125
Results
Lung cancer CT screening CAD systems’ performance has assessed for screening images of two clinics and Figure 2 and 3 show sample CAD detections.
According to Figure 2,
only four nodules from each lung has detected by the CAD system on Plaka screening clinics’ images of the selected location.
Therefore,
one false negative detection appears in each left and right lungs and the detection rate is 80%.
Figure 3 shows that all five nodules on left lung have detected while four from the right lung.
Therefore,
one false negative appears in the selected location on Azumi screening clinics’ image and the detection rate is 90%.
Furthermore,
the figures show that tested CAD system has performed better with clinic Azumi screening clinics’ images which were thick slices than thin slices from Plaka clinic.
FROC curves were plotted for CAD detections of both clinics separately.
Figure 4 demonstrates that CAD system has maximum TPF about 0.8 for thin slices and 0.9 for thick slices.
Therefore tested prototype CAD system has shown about 10% higher performance with Azumi images than Plaka images.