Localization of cardiac anatomical landmarks is an important step towards a more robust and accurate analysisof the heart. A fully automatic hybrid framework is proposed that detects key landmark locations in cardiacmagnetic resonance (MR) images. Our method is trained and evaluated for the detection of mitral valve pointson long-axis MRI and RV insert points in short-axis MRI. The framework incorporates four key modules for thelocalization of the landmark points. The first module crops the MR image around the heart using a convolutionalneural network (CNN). The second module employs a U-Net to obtain an efficient feature representation of thecardiac image, as well as detect a preliminary location of the landmark points. In the third module, the featurerepresentation of a cardiac image is processed with a Recurrent Neural Network (RNN). The RNN leverageseither spatial or temporal dynamics from neighboring slides in time or space and obtains a second predictionfor the landmark locations. In the last module the two predictions from the U-Net and RNN are combinedand final locations for the landmarks are extracted. The framework is separately trained and evaluated for thelocalization of each landmark, it achieves a final average error of 2.87 mm for the mitral valve points and anaverage error of 3.64 mm for the right ventricular insert points. Our method shows that the use of a recurrentneural network for the modeling of additional temporal or spatial dependencies improves localization accuracyand achieves promising results.
展开▼
