Hybrid UNet-Transformer model for Ultrasound Image Enhancement

Medical image enhancement is urgently needed to make healthcare systems more interpretable and diagnose more accurately. In standard deep learning architectures, it can be difficult to achieve the best of both worlds in terms of computational capability and efficiency of the feature extractor as well as the parameters. To that end, this study raises questions as follows: This study solves these problems by developing a hybrid UNet-Transformer model, which integrates Convolutional Neural Networks’ (CNNs) ability to capture localized spatial features and Transformers that can learn global context relations. This integration helps to segment and enhance images as well as possessing low computational complexity. To fine-tune the proposed model, hyperparameter sensitivity analysis in terms of learning rate, batch size and filter size is performed using ordinal parameter analysis. It should be noted that this analysis tries to serve as a guideline for refining the parameters with the aim of achieving better results. Hence, the effectiveness of this hybrid model is precisely tested using objective measures namely Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Mean Squared Error (MSE). This indicated that the proposed hybrid model yields outstanding performance as compared to other image enhancement techniques with PSNR=38.76, SSIM=98.6, MSE=0001.Interesting, the proposed hybrid image enhancement model can outperform other techniques. This further emphasizes the benefit of the model to retain key elements of the image while eliminating the noise in the image and enhancing the general quality of the image. This research presents a novel concept of feature extraction and parameter tuning that can be a base for establishing hybrid networks in medical image improvement. In this manner, the proposed methodology is beneficial in closing the gap between intricate recognition methods and real medical imaging implementations that serve to enhance diagnostic accuracy and speed in the medical field.