P2H-4 In Vivo Imaging of Coagulated Tissue

In this paper first results from a clinical study on imaging thermally coagulated tissue using a system for automated tissue characterization are presented. Complex baseband ultrasound data have been acquired during therapeutic treatments of liver tumors. The current study comprises 11 cases. For da...

Full description

Saved in:
Bibliographic Details
Main Authors: Siebers, S., Scheipers, U., Mohammad Ashfaq, Hansler, J., Frieser, M., Strobel, D., Hahn, E., Ermert, H.
Format: Conference Proceeding
Language:English
Subjects:
Attenuation
Baseband
Data acquisition
Image converters
In vivo
Liver neoplasms
Matrix converters
Radio frequency
Training data
Ultrasonic imaging
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper first results from a clinical study on imaging thermally coagulated tissue using a system for automated tissue characterization are presented. Complex baseband ultrasound data have been acquired during therapeutic treatments of liver tumors. The current study comprises 11 cases. For data acquisition, a conventional diagnostic ultrasound scanner was used. As a therapeutic device, a clinical approved radio-frequency (RF-) ablation device was chosen. Ultrasonic imaging was done prior to, during and after the treatment session intermittently. Acquired data were stored on an external PC and subdivided into numerous regions of interest (ROI). For each ROI, a set of tissue characterizing spectral and texture features was calculated. The best feature set was processed by the classification system. As training data, feature vectors have been calculated from ROIs representing definitely non coagulated and coagulated liver tissue. Classification was done by calculating maximum likelihood measures for each ROI. Likelihood maps hence obtained showed minima in actually coagulated areas. Starting with the minimum in the likelihood map, a region growing algorithm was used to threshold the maps and to convert them into binary images. Classification was done by total cross validation over cases. The best feature set was found by sequential forward selection and included attenuation, full width at half maximum, maximal correlation coefficient from cooccurrence matrices and gray level nonuniformity from gray level run length matrices. Sensitivity and specificity were determined as a quality measure of the classification results and yielded 0.87plusmn0.09 and 0.96plusmn0.03, respectively
ISSN:1051-0117
DOI:10.1109/ULTSYM.2006.443