Light-weight Frequency Re-calibration Network for Diabetic Retinopathy Multi-lesion Segmentation

Automated segmentation of Diabetic Retinopathy (DR) lesions is crucial for assessing DR severity and diagnosis. Most previous segmentation methods overlook the detrimental impact of texture information bias, resulting in suboptimal segmentation results. Additionally, the role of lesion shape is also not thoroughly considered. In this paper, we propose a light-weight frequency re-calibration network (LFRC-Net) for simultaneous multi-lesion DR segmentation, which integrates a Laplacian pyramid into the bottleneck layer of the encoder to analyze object features in different frequency domains. The high-frequency component handles texture information, while the low-frequency area focuses on learning the shape characteristics of DR lesions. By adaptively re-calibrating these frequency representations, our method can effectively differentiate the objects of interest. In the decoder, we introduce a residual attention module (RAM) to enhance lesion feature extraction and efficiently suppress irrelevant information. We evaluate the proposed model's segmentation performance on two public datasets, IDRiD and DDR, and a private dataset, Ultra-wide-field fundus images dataset. Extensive comparative experiments and ablation studies are conducted across multiple datasets. With minimal model parameters, our approach achieves mAP_PR of 60.51%, 34.83%, and 14.35% for the segmentation of EX, HE, and MA on the DDR dataset, and surpass those of the current state-of-the-art methods for EX and HE on the IDRiD dataset, which validates the effectiveness of our network.