Abstract
Accurate estimation of shape thickness from medical images is crucial in clinical applications. For example, the thickness of myocardium is one of the key to cardiac disease diagnosis. While mathematical models are available to obtain accurate dense thickness estimation, they suffer from heavy computational overhead due to iterative solvers. To this end, we propose novel methods for dense thickness estimation, including a fast solver that estimates thickness from binary annular shapes and an end-to-end network that estimates thickness directly from raw cardiac images. We test the proposed models on three cardiac datasets and one synthetic dataset, achieving impressive results and generalizability on all. Thickness estimation is performed without iterative solvers or manual correction, which is 100 times faster than the mathematical model. We also analyze thickness patterns on different cardiac pathologies with a standard clinical model and the results demonstrate the potential clinical value of our method for thickness based cardiac disease diagnosis.
Introduction
Measuring the thickness of the left ventricular (LV) wall is a cornerstone of cardiac disease diagnosis. Conditions such as hypertrophic cardiomyopathy (thickening) or dilated cardiomyopathy (thinning) are identified through these precise anatomical measurements. Currently, clinicians rely on two main approaches: Regional Wall Thickness (RWT), which averages thickness across zones, and Dense Wall Thickness (DWT), which calculates thickness at every point.
While DWT provides the necessary detail for localized pathology detection, it is traditionally calculated by solving Laplace’s equation to find the shortest path between the inner and outer heart walls. This mathematical approach is computationally intensive, often requiring iterative solvers that take a minute or more per volume and frequently necessitating manual intervention for complex contours. This research introduces a deep learning framework designed to provide real-time, automated dense thickness measurements.
Methodology
The researchers developed a two-stage approach to bridge the gap between raw pixels and anatomical measurements:
- The Fast Solver (U-Net-k): To eliminate the need for iterative solvers, the authors trained a U-Net to mimic the behavior of a Partial Differential Equation (PDE) solver. This model takes a binary mask of the heart shape and outputs a dense thickness map. To ensure robustness, it was trained on a synthetic dataset of 6,000 diverse annular shapes, teaching the model the underlying geometry of thickness regardless of image quality.
- The End-to-End Network (E2ET-solver): This model processes raw MRI images directly. It utilizes a shared encoder that branches into two tasks: segmentation (identifying the heart muscle) and thickness estimation.
A significant challenge in training such a network is "shape inconsistency": if a network predicts a slightly different heart shape than the ground truth, the original thickness map no longer aligns with the predicted shape. To solve this, the authors integrated their pre-trained Fast Solver into the training loop. It generates a "pseudo-ground truth" thickness map based on the predicted segmentation, ensuring the loss function compares consistent shapes.
Results and Clinical Utility
The proposed deep learning solver demonstrated significant improvements in both speed and accuracy:
- Efficiency: The model is 100 times faster than traditional mathematical models, reducing processing time from approximately 60 seconds to just 0.35 seconds per prediction.
- Accuracy: On the ACDC dataset, the E2ET-solver achieved a Mean Absolute Error (MAE) of 0.067 mm, outperforming baseline U-Net models (0.124 mm).
- Generalization: By learning geometric principles from synthetic data, the model generalized exceptionally well to unseen real-world datasets (LVQuan19 and MS-CMRSeg), proving more effective than models trained solely on real clinical data.
- Pathology Identification: The system successfully distinguished between different heart diseases, such as the distinct thinning patterns in Myocardial Infarction versus the thickening seen in Hypertrophic Cardiomyopathy.
Limitations and Future Directions
While highly effective, the model faces challenges near the cardiac apex due to the small region size and out-of-plane motion in 2D MRI slices. Future work aims to extend this framework to native 3D volumetric solvers to improve consistency across the entire heart volume and integrate the tool directly into MRI acquisition for real-time clinical feedback.
Summary updated