Cardiac magnetic resonance imaging（MRI） images precisely present anatomical structures and serve as a critical basis for diagnosing cardiovascular diseases.However

MRI images often suffer from inconsistent local details and global semantic representations

difficulties in maintaining spatial positioning and geometric continuity

as well as structural mis-segmentation and loss of lesion details.To address these issues

a cardiac MRI image segmentation network based on parallel receptive fields and attention mechanism is proposed.The method enhances the importance of regional features within the receptive field in the spatial domain

while enabling smaller convolutional kernels to capture a larger receptive field in the frequency domain. Additionally

a positional-interpolated gating attention is designed to generate positional encodings to preserve geometric consistency

and a gating unit is used to modulate the attention output features.Finally

a feature interaction fusion module is adopted to eliminate dimensional and semantic discrepancies between local features and global semantics.Experiments on the public ACDC and MSCMRseg datasets demonstrate that the proposed method achieves higher segmentation accuracy than existing cardiac MRI image segmentation methods.