Empirical mode decomposition and Hessian LLE in Fluorescence spectral signal analysis for Cervical cancer detection

Cervical cancer is a significant cause of female mortality worldwide. The timely and accurate identification of different stages of cervical cancer has the capacity to significantly improve both treatment efficacy and patient survival duration. Fluorescence spectroscopy acts as a significantly sensitive technique for identifying the biochemical changes that occur during the advancement of cancer. Fluorescence spectral data was collected from a diverse set of 110 human cervix samples in our study. The spectral data underwent an initial preprocessing step that included data normalization. Subsequently, empirical mode decomposition (EMD) was utilized to decompose the signal into several intrinsic mode functions within the spectral domain. Thereafter, various nonlinear dimensionality reduction methods, including Isomap, Local Linear Embedding (LLE), and Hessian LLE, were applied to extract more informative features in a lower-dimensional representation. Furthermore, a 1D convolutional neural network (CNN) was employed to categorize the lower dimensional spectral signals into three classes: normal, pre-cancerous, and cancerous. The proposed methodology attained the best evaluation metrics using the Hessian LLE dimensionality reduction technique. A mean classification accuracy, precision, recall, F1-score, and specificity of 98.72%, 98.02%, 98.61%, 98.00%, and 99.30%, respectively, were obtained through a 5-fold cross-validation technique. The combination of fluorescence spectroscopy and machine learning holds promise for detecting cancer at earlier stages than current diagnostic methods.