At present, Optical Coherence Tomography Angiography (OCTA) is more and more widely used in the diagnosis of fundus diseases. Compared to other imaging methods, OCTA provides non-invasive, high-resolution imaging of blood vessels. The circular avascular area formed by the retinal vascular bed is known as the Foveal Avascular Zone (FAZ). Various retinal diseases, such as diabetic retinopathy, can lead to capillary ischemia, where the ischemic capillaries cannot be imaged in OCTA, causing changes in the shape and size of the FAZ. Therefore, FAZ segmentation is crucial for the diagnosis of retinal diseases. FAZ segmentation faces three main challenges: low contrast and high noise, varying sizes and shapes, and protrusions or groovs caused by lesions. This study proposes a ConvNeXt block based deep learning model with U-shaped structure and a weighted Dice loss to specifically address these three challenges. Therefore, The proposed approach is called as edge-enhanced ConvNeXt-Unet (EE-ConvUnet), and it can pay more attention to the boundary area of FAZ. The proposed method is validated on a public dataset sOCTA-3 \(\times \) 3-1.1k-seg, an average Dice of 0.932 and an average Intersection over Union (IoU) of 0.877 are obtained. These results demonstrate the effectiveness of the proposed method in FAZ segmentation.