🤖 AI Summary
Accurate and robust automatic segmentation of the left ventricle in cardiac MRI is crucial for clinical diagnosis, yet existing methods still face challenges in precision and generalizability. This work proposes two novel U-Net variants, LNU-Net and IBU-Net, which incorporate layer normalization (LN) and instance-batch normalization (IBN), respectively, to enhance model stability and performance. To further improve generalization, the training pipeline integrates data augmentation through affine transformations and elastic deformations. Evaluated on a dataset comprising 45 patients and 805 short-axis cine MRI slices, the proposed methods consistently outperform current state-of-the-art approaches in key metrics, including the Dice coefficient and mean perpendicular distance, demonstrating their effectiveness and technical innovation in left ventricular segmentation.
📝 Abstract
Left ventricle (LV) segmentation is critical for clinical quantification and diagnosis of cardiac images. In this work, we propose two novel deep learning architectures called LNU-Net and IBU-Net for left ventricle segmentation from short-axis cine MRI images. LNU-Net is derived from layer normalization (LN) U-Net architecture, while IBU-Net is derived from the instance-batch normalized (IB) U-Net for medical image segmentation. The architectures of LNU-Net and IBU-Net have a down-sampling path for feature extraction and an up-sampling path for precise localization. We use the original U-Net as the basic segmentation approach and compared it with our proposed architectures. Both LNU-Net and IBU-Net have left ventricle segmentation methods: LNU-Net applies layer normalization in each convolutional block, while IBU-Net incorporates instance and batch normalization together in the first convolutional block and passes its result to the next layer. Our method incorporates affine transformations and elastic deformations for image data processing. Our dataset that contains 805 MRI images regarding the left ventricle from 45 patients is used for evaluation. We experimentally evaluate the results of the proposed approaches outperforming the dice coefficient and the average perpendicular distance than other state-of-the-art approaches.