Hybrid CNN-LSTM Transfer Learning for Dengue Diagnosis from Raman Spectroscopy Images

The diagnosis of dengue fever can be made quickly and effectively with Raman spectroscopy. However, the diagnostic procedure presents difficulties; for example, the distinction between healthy and infected spectra is extremely subtle. Deep learning (DL)-based diagnosis of dengue from Raman spectroscopy is gaining research interest. This paper introduces an innovative hybrid model that combines the architecture of a long-short-term memory recurrent neural network (LSTM RNN) with the capabilities of a pre-trained convolutional neural network (CNN). We present the CNN-LSTM Transfer Learning (HCL-TL) method, which uses Raman spectroscopy images to transform the field of data-driven dengue disease diagnosis. The proposed approach utilizes hyperparameters to refine the DL models. Both the initial training and subsequent fine-tuning stages incorporate model checkpoints and callbacks to monitor loss and prevent overfitting. A dataset consisting of 635 spectra-460 of which were infected and 175 of which were healthy-was used to evaluate the suggested model. Both holdout and 10-fold cross-validation methods are used for splitting the data samples. When applied to testing data for dengue infection diagnosis, the proposed HCL- TL method achieves a classification accuracy of 97.73%. The HCL-TL method outperforms the state-of-the-art DL and machine learning models in terms of accurately classifying dengue fever from Raman spectroscopy images, according to the results.