Automatic segmentation of ablation lesions and termination of the image acquisition/analysis process
- PDF / 1,471,871 Bytes
- 3 Pages / 595.276 x 793.701 pts Page_size
- 36 Downloads / 184 Views
POSTER PRESENTATION
Open Access
Automatic segmentation of ablation lesions and termination of the image acquisition/analysis process Andriy V Shmatukha1*, Eugene Crystal2 From 2011 SCMR/Euro CMR Joint Scientific Sessions Nice, France. 3-6 February 2011 Objective Automate the processes of ablation lesion imaging and delineation in order to make them non-expert user friendly. Background Visualization of radiofrequency ablation lesions during cardiac electrophysiology procedures would help ensuring their contiguity and inclusiveness, which are essential for the procedures’ long-term success. The usefulness of dynamic contrast enhancement (DynCE) and cumulative characteristics for ablation lesion visualization has been already demonstrated (1). However, planning of the image acquisition process and interpretation of the resulting images may pose a challenge for electrophysiologists who don’t interpret MRI routinely. We describe an algorithm allowing automatic discrimination between ablation lesions and surrounding normal tissue during DynCE scans as well automatic termination of the image acquisition and analysis processes as soon as the desired lesion visibility level has been achieved. Methods 56 lesions were ablated in the Latissimus dorsi muscles of 15 rabbits using clinical catheters and time/power settings. The animals underwent MRI at various times after ablations using various imaging techniques. DynCE images were post-processed using original algorithms and software (1).
1 Cardiac and Interventional Applied Science Laboratory, General Electric Healthcare, Toronto, ON, Canada Full list of author information is available at the end of the article
Results Lesion non-detectability on early contrast agent wash-in cumulative DynCE images strongly correlated with lack of lesion and normal tissue separation on their histograms (Fig. 1). As wash-in continued and new data was acquired and post-processed, ablation lesions became more apparent (Fig. 2) and separated from normal tissue on histograms (Fig. 3): lower-intensity histogram peaks were formed by lesion core pixels, higher-intensity peaks were formed by normal tissue pixels, and lesion border pixels composed the groove segment between these peaks (Fig. 4). Our algorithm automatically identified the peaks and groove, and used the information to discriminate between actively and poorly enhancing pixels (Fig. 5). It also compared the peaks’ values to the groove’s one and used the information to terminate image post-processing when satisfactory lesion-to-tissue contrast was detected (Fig. 6). The resulting segmented images demonstrated good correspondence to other lesion depicting MR images acquired during the study (Fig. 7). Conclusions Our algorithm demonstrated a good performance in this study and has a potential to prove robust and useful in real clinical conditions. More accurate and noise-resistant histogram analysis and segmentation methods can be implemented, which would result in more robust ablation lesion delineation and the reduction of the DynCE scan
Data Loading...
