Repair or Replace? Answering a Surgical Dilemma with an Automated Image Analysis Tool
Ahmed H. Aly1, Frank Meijerink2, Thomas J. Eperjesi3, Wobbe Bouma4, Joseph H. Gorman, III5, Paul A. Yushkevich6, Robert C. Gorman5, Alison M. Pouch6.
1Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA, 2University of Groningen, Groningen, Netherlands, 3Gorman Cardiovascular Research Group, University of Pennsylvania, Philadelphia, PA, USA, 4Department of Cardiothoracic Surgery, University of Groningen, Groningen, Netherlands, 5Gorman Cardiovascular Research Group, Department of Surgery, University of Pennsylvania, Philadelphia, PA, USA, 6Department of Radiology, University of Pennsylvania, Philadelphia, PA, USA.
OBJECTIVE:Severe ischemic mitral regurgitation (IMR) presents a critical dilemma in mitral valve (MV) surgery decision making. The first randomized control trial conducted by the Cardiothoracic Surgery Network (Acker et al NEJM 2014 and Goldstein et al NEJM 2016) showed that a subset of MV repair with annuloplasty group that did not recur had the best outcomes. Overall, MV repair had a recurrence rate of 60% compared to 4% in replacement. The problem of recurrence could be reduced significantly if MV repair were only performed on patients likely to have a durable outcome. Our group (Bouma et al Ann Thorac Surg 2016) showed that P3 tethering angles can serve as a strong predictor of IMR recurrence after MV repair (AUC = 0.92). While these results are promising, manual tracing of MV 3D transesophageal echocardiography (TEE) images is laborious and not practical for the operating room (OR) environment. The objective is to develop an automated image analysis tool for 3D modeling of the MV with automatic tethering angle and feature extraction.
METHODS: 3D TEE pre-operative images (N = 66) were obtained at the Hospital of the University of Pennsylvania from 46 MV repair patients (12 recurrent and 34 non-recurrent) and 20 controls. MV models were produced as described in Pouch et al Med. Image. Analysis 2014. MV tethering angles were extracted automatically using MATLAB.
RESULTS:The automatically extracted P3 tethering angles show significant difference between nonrecurrent and recurrent and between normal and recurrent. The P3 tethering angle p-values in Table 2 are comparable with the results based on manual segmentation (p<0.01).
|Table 1: Segmental tethering angle measurements|
|Normal (n = 20)||Nonrecurrent (n = 34)||Recurrent (n = 12)|
|Tethering angle (degrees)||avg||std||avg||std||avg||std|
|Table 2: p-values of t-test comparison of tethering angles|
|comparison / tethering angle||A1||A2||A3||P1||P2||P3|
|normal vs nonrecurrent||0.4988||0.0891||0.6920||0.9883||0.1831||0.4802|
|normal vs recurrent||0.2216||0.1977||0.3555||0.9398||0.0233||0.0015|
|nonrecurrent vs recurrent||0.4280||0.9183||0.5722||0.9383||0.2762||0.0057|
CONCLUSIONS: The automated extraction of features such as P3 tethering angles from MV models is a promising tool for pre-operative IMR recurrence prediction.Evaluation on a larger data set will be done before conducting a feasibility study in the OR.
Back to 2018 Program