Research on Retinal Vascular Pathologies in OCT Images with the Coupled ResNet and Transformer Model

Retinal vascular pathologies can lead to vision loss, blindness, and severe complications (such as glaucoma and vitreous hemorrhage), significantly impacting quality of life. Traditional detection methods for retinal vascular pathologies rely on manual observation and subjective judgment, resulting in low efficiency and a high risk of missing early subtle lesions. They also suffer from limitations such as operational complexity and device dependency. To overcome the shortcomings of these methods, this study proposes a method for detecting retinal pathologies on the basis of OCT images of retinal vascular diseases that uses a coupled ResNet and transformer approach. RTHNet adopts an encoder-decoder structure. In the encoding stage, ResNet50 is employed as the backbone network to extract local features effectively from the images. An attention adaptive fusion module (AAFM) is designed to achieve efficient integration of multilevel attention features between the encoder and decoder. In the decoding stage, a global-local contextual transformer block (GLCTB) is designed to simultaneously focus on global contextual information and local details. At the end of the decoder, a detail enhancement module (DEM) is proposed, which refines the semantic consistency and spatial detail information between features to ensure the fineness and accuracy of the segmentation results. The RTHNet model was evaluated on the OCT2017 dataset and OCTAMNIST dataset. The results revealed that the classification accuracy reached 94.78% and 83.41% on the retinal OCT2017 dataset and OCTAMNIST dataset, respectively. Compared with other traditional methods, such as the CNN and DCNN algorithms, the detection accuracy improved by 7.28% and 0.78%, respectively. The proposed method, which couples ResNet and Transformer, overcomes the bottlenecks of subjectivity and low efficiency inherent in traditional detection methods that rely on manual interpretation. It enables high-precision automatic identification and early warning of vascular pathologies in OCT images, providing an intelligent auxiliary diagnostic tool for clinical practice. This reduces misdiagnosis rates and promotes the automation and precision of ophthalmic disease screening.