Mariam Afshin, Presenter: Nothing to Disclose

Ismail Ben Ayed, Abstract Co-Author: Nothing to Disclose

Hamid Reza Sadeghi Neshat MSc, Abstract Co-Author: Nothing to Disclose

Kumaradevan Punithakumar, Abstract Co-Author: Nothing to Disclose

Aashish Goela MD, Abstract Co-Author: Nothing to Disclose

Ali Islam MD, Abstract Co-Author: Nothing to Disclose

Terry Peters, Abstract Co-Author: Nothing to Disclose

Shuo Li PhD, Abstract Co-Author: Employee, General Electric Company

We developed a regional cardiac abnormality detection framework based on the intensity features in MR images estimated from user-provided delineation of three main slices of the first frame. The algorithm was tested on 90×20 LV cavities of short-axis MR images of 30 subjects and was demonstrated as a promising diagnostic tool with more than 90% overall accuracy.

Early and accurate detection of Left Ventricle (LV) regional wall motion abnormalities significantly helps in the diagnosis and follow-up of cardiovascular diseases. In routine clinical use, cardiac function is estimated by visual assessment and interpretation of LV. Therefore, it is highly subject-dependent. We have developed a regional myocardial abnormality detection framework using MR Image statistics with minimal user input.

440 out of the total 480 regional segments were successfully diagnosed. The abnormalities in apical slices were diagnosed with 92.8% accuracy, compared to 91.3% for midcavity and 90.5% for basal slices. We believe the reason is due to the absence of papillary muscle in apical slices making the intensity features more distinguishable.

Following IRB approval, we collected cardiac cine MR images of patients using 1.5T GE scanner with FIESTA image sequence.For each patient, the user was only asked to provide manual delineation of the LV and two pixels on the septal wall of the first frame (in three slices, basal, mid-cavity and apical). Then, the software automatically estimated the 16 regional segments of this frame following American Heart Association (AHA) standards. Next, the estimated segments were systematically superimposed to all other frames. We then estimated image intensity statistics corresponding to each regional segment and used it as a feature to determine whether each regional segment function is normal or not using a nonlinear classification technique.We tested our algorithm on 30 subjects, 90 short-axis image sequences (apical, mid-cavity and basal slices of 30 patients), Each sequence consisted of 20 functional 2D images. The results for 480 myocardial segments from the three slices were compared with ground truth classifications by two experienced radiologists.

Afshin, M, Ben Ayed, I, Sadeghi Neshat, H, Punithakumar, K, Goela, A, Islam, A, Peters, T, Li, S, Automated Assessment of Regional Left Ventricular Function for Cardiac MRI with Minimal User Interaction.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12026909.html