Diagnosis of ocular tumor diseases based on serum Raman spectroscopy combined with AE-LSTM network

Ocular tumors are a common type of tumor and accurate diagnosis of ocular tumors is critical for effective treatment and prognosis. Raman spectroscopy is being used as a more efficient and non-invasive method for disease diagnosis. Due to their strong feature extraction capabilities, deep learning models have become a promising approach for disease classification tasks involving Raman spectroscopy. However, the original models have limitations in using Raman spectroscopy data for disease prediction, which cannot effectively utilize the temporal features of Raman spectroscopy and pay more attention to the significantly changing feature peaks. This study introduces an AE-LSTM architecture that incorporates a self-attention mechanism into a long short-term memory framework, aiming to enhance the extraction of both global context and local details from Raman spectral signals. The experimental results show that the model is able to get the correct prediction of ocular tumors and healthy tissues with an accuracy of 98.46%. Compared with the traditional machine learning model and other deep learning methods, all the indexes reach the optimum. The demonstrated efficacy of this method in studying ocular tumors offers a novel framework for extending spectral analysis coupled with deep learning architectures to diverse biomedical applications.