HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH SEARCH RESULT		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Petrocelli, R. Articles by Yahil, A. Search for Related Content PubMed Articles by Petrocelli, R. Articles by Yahil, A.

Improvement in diagnostic performance and reader confidence level enabled by Pixon image processing of planar nuclear studies

¹ Geisinger Medical Center, Danville, Pennsylvania ² Siemens Medical Solutions USA, Hoffman Estates, Illinois ³ Pixon LLC, Stony Brook, New York

1415

Objectives: Routine clinical planar nuclear studies still lack sophisticated, statistics-based image processing, although the planar studies make up a large fraction of the workload. The Pixon method uses adaptive denoising and deblurring to improve image quality. We investigate the clinical impact of Pixon processing on planar images in a variety of cases by measuring the preference of readers for the processed images and their score of the improvement in diagnostic performance and reader confidence level of processed versus unprocessed images.

Methods: 7 physicians with no previous experience with Pixon processing are presented with 29 cases (1 abscess, 8 bone, 2 lymphatic, 1 Ceretec WBC, 2 CSF leaks, 1 GI bleed, 4 I-131, 2 thyroid, 4 parathyroid, 2 V/Q, 1 voiding cystogram). The cases are a random subset selected by RDP, an experienced NM physician, from challenging cases obtained at the Geisinger Medical Center (Danville, PA) over a 3-month period. The original and processed images are presented to each reader side by side, using custom software that randomizes the order of the cases and allows the color table and display windowing to be adjusted by the reader. The processed images presented are fusion images consisting of reader-selectable linear mixtures of the unprocessed images and the pure Pixon-processed images. For each image, the reader first selects the preferred percentage of processed image in the fusion, with 0%=original image and 100%=pure processed image. The reader then scores the improvement in diagnostic performance and confidence on a 5-point scale: -2=significant deterioration, -1=some deterioration, 0=no change, +1=some improvement, +2=significant improvement.

Results: The median mixing percentage selected in all 203 samples (29 cases x 7 readers) is 90% and the average 74%, a clear preference for processed images. The distribution of diagnostic performance scores among the 203 samples is +2: 15 (7%), +1: 60 (30%), 0: 121 (60%), -1: 6 (3%), -2: 1 (0%). The distribution of diagnostic confidence scores is +2: 30 (15%), +1: 86 (42%), 0: 80 (39%), -1: 7 (3%), -2: 0 (0%). Thus, the readers find improvement in diagnostic performance and confidence in 37% and 47% of the samples and no change in 60% and 39% of the samples, respectively, a total of 97% for both diagnostic performance and confidence.

Conclusions: The readers show clear preference for the processed images by selecting high mixing percentages of the processed images and consider the diagnostic performance and confidence of the processed image to be equal to or better than the original images in 97% of the samples.