Phase Analysis of Radionuclide Ventriculography: Assessment of Herceptin Cardiotoxicity Amna Juma Al-Jabri1,2, Craig Paterson2,3, Jamie Robinson2,3, Bill Martin3 , Scott Smith3 , Sandra Reid3 1Department of Clinical Physics and Bioengineering, NHS GG&C, 2School of Medicine, University of Glasgow, 3Department of Nuclear Cardiology, NHS GG&C, Introduction and Aim Herceptin is a human epidermal growth factor receptor 2 (HER2) targeted monoclonal antibody with a proven efficacy in the treatment of HER2+ metatstatic breast cancers, when combined with chemotherapy. However, Herceptin therapy carries a significant risk of cardiotoxicity, which can manifest as an asymptomatic decrease in left ventricular ejection fraction (LVEF) and heart failure. Consequently, serial assessment of left ventricular function is imperative whilst patients are undergoing Herceptin therapy. This is achieved by using Radionuclide Ventriculography (RNVG) to monitor LVEF. However, a drop in LVEF is a late sign of cardiac damage, which may be irreversible. This technique is also sensitive to slight changes in cardiac function. The aim of this work was to investigate whether established RNVG phase parameters such as Entropy, Synchrony, Approximate Entropy and Symbolic Entropy may be used as an alternative, more sensitive predictor of cardiac damage. The phase parameterscan be extracted via Fourier analysis of RNVG data (Fig. 1). The work also investigated the serial changes in the phase parameters over the course of Herceptin therapy.   Entropy in Symbolic Dynamics Symbolic dynamics is a means of coarse graining a system. It is used to model a smoothly varying dynamic system by discrete sequences of abstract symbols. Applying symbolic dynamics to RNVG data is achieved by dividing the phase image into segments, where each pixel is assigned a segment dependent upon how far its phase value is from the start of ventricular contraction. Each segment in the plot is then assigned an associated symbol which represents how far out of phase the pixel is. This creates a string of symbols which can be analyzed to determine the frequency with which patterns (words) of varying length occur. The frequency information can then be used to calculate the Shannon entropy for that frequency distribution. Symbolic dynamic entropy has previously been used at GRI to characterize heart rate (R-R) variability in patients with atrial fibrillation. It has been shown to offer an increase in the complexity of the measure of entropy (as compared with Shannon entropy) by establishing a well defined set of states for the system.   Results Fig. 1. Amplitude (left) and phase (right) images extracted from a Fourier analysis of RNVG data. The atria are displayed out of phase with the ventricles and the right ventricle displays a Right Bundle Branch Block. Matlab was used to extract phase parameters for all 36 serial Herceptin datasets. A phase histogram for a normal study is displayed in Fig. 2 showing the normal phase distributions for both the left and right ventricles. Once the phase parameters were extracted Receiver Operator Characteristic (ROC) curves were generated using the LVEF assessment of change in cardiac function as a gold standard. Fig. 3 shows the ROC curve generated for each parameter for images acquired after the commencement of Herceptin therapy. As can be seen the highest area under the ROC curve (AUC) is for the measures of entropy and approximate/symbolic entropy. Table 1. lists the AUC values for each parameter at each stage during therapy. Entropy and Approximate Entropy consistently give the largest AUC values. Method Current protocol at the Western Infirmary, Glasgow, is for Herceptin patients to receive a baseline frame mode RNVG investigation before commencing Hercetpin and chemotherapy. Further imaging is then performed every three months during therapy to monitor for any changes in cardiac function. Therapy is stopped if there is a greater than 10% drop in LVEF from baseline or if LVEF falls below 50%. RNVG imaging is performed by in vivo labelling of red blood cells with 600 MBq of 99mTc-Pertechnetate. 2 ml of Pyrophosphate is administered to the patient 20 minutes before administration of the 99mTc. The PYP binds to the red blood cells and prevents the 99mTc from diffusing back out leading to 85% - 95% labeling efficiency. A 3-lead monitoring quality ECG is recorded and the upslope of the R-wave is used for gating of the data To extract phase information from the RNVG data time activity curves (TACs) are generated by plotting intensity versus time for each ventricular region of interest. Phase and amplitude information are then extracted from the data by fitting first and second order Fourier harmonics to the TACs. Fig. 2 displays a phase histogram for normal Ventricles. Phase information was extracted for serial Assessments in 36 patients (17 normal and 19 abnormal; as determined by LVEF assessment). The parameters of Entropy, Synchrony, Approximate Entropy and Symbolic Entropy were calculated for each of the serial studies. Parameter Baseline 3 Month Scan 6 Month Scan 9 Month Scan Synchrony 0.69 0.52 0.65 0.67 Entropy 0.89 0.79 0.86 0.87 App Entropy 0.82 0.74 0.81 Symbolic Entropy 0.80 0.72 0.85 LVEF 0.50 0.51 0.42 0.34 Sensitivity Tab. 1. AUC values for each parameter at baseline and at 3 month intervals post start of Herceptin. Entropy and Approximate/Symbolic Entropy consistently have the largest AUC values throughout therapy. 1 - Specificity Fig. 3. ROC curves for each phase parameter at start of therapy. Boxplots were generated for each parameter for every scan in the serial assessment. A normal range was generated for each parameter where previous LVEF results were used to determine if a study was normal. The boxplots displayed in Fig. 4. demonstrate how the parameters change over time for both normal and abnormal studies. The plots also display the total number of patients included at each study time. Only Entropy and Approximate Entropy are able to differentiate between the normal and abnormal groups (p = 0.003 and p = 0.03 respectively for 1st study after therapy starts). Significant difference between the two datasets can seen as soon as therapy starts and at each subsequent imaging point. Fig. 2. Phase histograms for a normal synchronous patient (left) and a patient with RBBB (right)   Fig. 4. Boxplots for Synchrony, Entropy, Approximate Entropy and LVEF over the time course of Herceptin therapy. Conclusion ROC curves were generated for each phase parameter and results were compared with the standard measure of LVEF. The AUC was highest for Entropy, Symbolic Entropy and Approximate Entropy (> 0.82), while for synchrony it was 0.69 and LVEF 0.51. Throughout the course of Herceptin therapy the entropy parameters had significantly higher AUC values than synchrony and LVEF, and gave the greatest differentiation between normal and abnormal studies. The study demonstrated that novel phase parameters of Entropy and Approximate/Symbolic Entropy have improved diagnostic accuracy for detection/prediction of cardiotoxic effects – as compared to the currently used measure of LVEF. The measure of cardiac phase Synchrony produced similar results to LVEF. A large scale prospective study is required to fully evaluate the potential use of these new parameters in assessment of Herceptin cardiotoxicity. Eqn. 1. Synchrony Eqn. 2. Entropy References [1]: O’Connell J. William, et al, Journal of Nuclear Cardiology, 2005, 12, p 441-450 [2]: Cullen J, et al, Biomedical Signal Processing and Controls, Vol. 5, Issue. 1, 2010, p 32-36 Eqn. 3. Approximate Entropy