High-Precision Lung Cancer Localization Precision with Histogram Equalisation and Frequency-Domain Hybrid Attention

The advent of deep learning algorithms, in particular Convolutional Neural Networks, has provided robust technical support for medical image processing. We propose solutions to the challenges of poor image quality, insufficient model accuracy, and lost frequency domain information. These solutions include a histogram equalisation module, a gated feature selection mechanism, and a hybrid frequency-domain attention module. The integration of these three modules into the U-Net architecture has been demonstrated to enhance performance through the implementation of feature filtering and fine-grained frequency domain capture. Compared with the baseline U-Net model, the Histogram Enhancement–Fourier Transform & Laplace Transform Attention U-Net (HFLU-Net) improved the accuracy metric for localising small-scale lung cancer lesions to 0.6889. This finding indicates that the model contributes to the enhancement of radiotherapy precision for patients with early-stage lung cancer.