Synopsis

Keywords: Machine Learning/Artificial Intelligence, Segmentation

We proposed and validated Data Adaptive Uncertainty-Guided Spatiotemporal (DAUGS) analysis that leverages the data-driven uncertainty map of the segmentation contours among a pool of trained deep neural networks (DNNs) and automatically selects the segmentation result with the highest level of certainty. Our results suggest that proposed DAUGS and standard DNN-based analysis demonstrated on-par performance on the internal test set which is from the same institution as training set and acquired with FLASH sequence. In contrast, DAUGS analysis considerably outperformed DNN-based analysis on the external test set which was acquired with a bSSFP pulse sequence at a different institution, demonstrating the improved robustness of the proposed method despite limited training data.

Introduction

Fully automatic analysis of first-pass perfusion (FPP) myocardial MR datasets enables rapid and objective reporting of stress/rest studies in patients with suspected ischemic heart disease [1-3]. With deep learning-based approaches, training well-generalizable deep neural network (DNN) models that, despite having a limited training dataset, are robust across different sites and data-acquisition protocols is an ongoing challenge. Previous work has shown that a “sliding patch” approach for analysis of FPP images generates a data-driven pixelwise “uncertainty map” as a byproduct of the segmentation process [4,5].

Methods

We propose to leverage the data-driven uncertainty map (U-map) among a pool of multiple trained DNNs (all with the same 3D U-Net architecture but trained with different parameter initializations) to perform Data Adaptive Uncertainty-Guided Spatiotemporal (DAUGS) analysis as in Fig 1, and automatically select the “best” segmentation result with the highest level of certainty in a data-adaptive manner by computing total per-pixel energy U_pp of the U-maps (Step 4 in Fig 1). FPP data from 106 patients with suspected ischemia and 14 healthy subjects acquired at 3T from two sites were used: (1) an internal dataset acquired using a saturation recovery (SR) prepared FLASH sequence, and (2) an external dataset acquired with SR-prepared SSFP (Fig 2). Training of DNNs used the data from a subset of the internal dataset (330 stress/rest FPP image series; 90% females). Performance of the proposed DAUGS vs. standard DNN-based analysis was evaluated on a small subset of the internal dataset which has no overlap with the training data and the entire external dataset (120 stress FPP image series; 25% females).

Results

Fig 3 summarizes the Dice-score comparison of the proposed DAUGS analysis approach vs. standard DNN-based analysis. The “standard” DNN-based analysis approach refers to the conventional DNN training approach in which a single model is selected during the validation process. For the internal dataset, our proposed method and standard approach showed a comparable performance (p >0.5). However, on the external dataset, ours significantly outperformed the standard approach (Dice: 0.885 ± 0.040 vs. 0.849 ± 0.065, p < 0.01). Fig 4 shows a challenging case from the external test set with diffuse stress-induced ischemia in all 3 short-axis slices and LV hypertrophy which renders the mid slice (acquired end systole) very challenging to segment. The segmentation chosen by the proposed approach performs well with a mean Dice score of >0.90, whereas the standard approach fails to segment to mid slice. Overall, the number of “failed” segmentations (discontiguous contours) was markedly lower for the proposed method (< 1% vs. 5%).

Conclusion

Our proposed data-adaptive approach for analysis of FPP datasets offers the flexibility to choose the final segmentation result from a pool of candidate solutions based on the uncertainty level detected by the trained spatiotemporal DNNs. Our results demonstrate that the proposed DAUGS analysis approach improves the generalization ability of DNN-based analysis despite the limited training data, which in turn has the potential to enable automatic analysis of perfusion CMR datasets with improved robustness to variations in the data acquisition protocol (SR- FLASH vs. SR-SSFP), sequence parameters, or site location.

Acknowledgements

No acknowledgement found.

References

1. Xue H, Davies RH, Brown LA, Knott KD, Kotecha T, Fontana M, et al. Automated inline analysis of myocardial perfusion MRI with deep learning. Radiology: AI 2020;2(6):e200009.

2. Scannell CM, Veta M, Villa ADM, Sammut EC, Lee J, Breeuwer M, et al. Deep-Learning-Based Preprocessing for Quantitative Myocardial Perfusion MRI. JMRI 2020;51(6):1689-96.

3. Hann E, Popescu IA, Zhang Q, Gonzales RA, Barutçu A, Neubauer S, et al. Deep neural network ensemble for on-the-fly quality control- driven segmentation of cardiac MRI T1 mapping. Medical image analysis. 2021 Jul 1;71:102029.

4. Yalcinkaya DM, Youssef K, Heydari B, Zamudio L, Dharmakumar R, Sharif B. Deep Learning-Based Segmentation and Uncertainty Assessment for Automated Analysis of Myocardial Perfusion MRI Datasets Using Patch-Level Training and Advanced Data Augmentation. Proceedings of IEEE Engineering in Medicine & Biology Society (EMBC); 2021;4072-78.

5. Lu MY, Williamson DF, Chen TY, Chen RJ, Barbieri M, Mahmood F. Data-efficient and weakly supervised computational pathology on whole-slide images. Nature Biomedical Engineering 2021;5(6):555-70.