Premium
SU‐FF‐I‐12: Boundary Based Vs. Region‐Based Segmentation Techniques for Breast Lesion Phantoms Produced by Fischer's Full Field Digital Mammography Ultrasound System (FFDMUS): A Novel Tool for Performance Evaluation of LCD's and CRT's
Author(s) -
Suri J,
Poolla A,
Ye Z,
Bilhanan A
Publication year - 2005
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.1997492
Subject(s) - segmentation , ultrasound , artificial intelligence , standard deviation , computer science , computer vision , breast ultrasound , mammography , digital mammography , mathematics , medicine , radiology , breast cancer , statistics , cancer
Purpose: This paper presents a performance evaluation strategy for LCD and CRT display characterization using ultrasound data acquired from Fischer's full field digital mammography and ultrasound system (FFDMUS) prototype. Method and Materials: Lesion segmentation of these ultrasound images acquired through the FFDMUS was performed using two approaches: (a) Gradient Vector Flow (GVF) and (b) Signal‐to‐Noise (SNR). The protocol consisted of FFDMUS ultrasound data acquisition with known X‐ray and ultrasound parameters. The ultrasound ROI images were displayed on LCD and CRT displays, memory grabbed and then segmented using GVF and SNR tools. Note that during the memory grabbing process from the displays, the spatial properties of the displays were ignored. This output was then optimized for partial volume correction, given the ideal boundary. The performance of the segmentation algorithms was evaluated by quantifying the mean error between the ideal boundary and the computer‐estimated boundary. Polyline distance metric (PDM) was used as a ruler. We used the segmentation and error quantification system on LCD and CRT displays. Results: we optimized our segmentation algorithm over 22 ultrasound FFDMUS images. The mean and standard deviation using GVF on LCD/CRT were 0.870 & 0.207/0.900 & 0.244. Correspondingly, the mean and standard deviation using SNR on LCD/CRT were 1.129 & 0.321/1.105 & 0.315. Conclusion: It was observed that both the segmentation strategies are comparable on both LCD's and CRT's for this data set used. Also the GVF performed better than SNR, as it was an interactive methodology.