Enhancing Terahertz Spectral Recognition of Lung Cancer Cells Through Synthetic Signal Generation

Deep learning-based medical diagnosis models heavily rely on large-scale datasets encompassing diverse aspects of a patient’s condition. However, the scarcity of such datasets, due to challenges in acquiring biological samples, presents a significant limitation. Synthetic data generation offers a promising solution to advance public healthcare while preserving patient confidentiality, and eliminating the need for physical samples. Nonetheless, existing synthetic data generation methods often suffer from limited diversity and quantity. This paper introduces EDGAN(Encoder-Decoder Generative Adversarial Network), a novel generative adversarial network (GAN) model specifically designed for the automated generation of terahertz (THz) signals in lung cancer samples. The proposed EDGAN model adopts a convolutional neural network (CNN) architecture, incorporating an encoder-decoder generator network and a discriminator network. Through comprehensive evaluation using multiple metrics, our study demonstrates the superior performance of the EDGAN model compared to other GAN models. Notably, the synthetic signals generated by our approach exhibit exceptional accuracy when assessed using alternative classification models, underscoring its potential in medical diagnosis applications.