Optimal Scree-CNN for Detecting NS1 Molecular Fingerprint from Salivary SERS Spectra

Area of Science:

• Biomedical Engineering
• Machine Learning
• Spectroscopy

Background:

• Dengue Fever (DF) is a serious viral infection.
• Current diagnostic methods rely on serum and antibody detection.
• NS1 antigen in saliva presents a potential noninvasive biomarker for early DF detection.

Purpose of the Study:

• To develop and optimize a Scree-Convolutional Neural Network (CNN) model for classifying salivary NS1 Surface-Enhanced Raman Spectroscopy (SERS) spectra.
• To evaluate the performance of the Scree-CNN model against established diagnostic standards for Dengue Fever.

Main Methods:

• Utilized Surface-Enhanced Raman Spectroscopy (SERS) to capture NS1 molecular fingerprints in saliva samples (284 samples, 1801 features each).
• Applied Principal Component Analysis (PCA) for dimensionality reduction of spectral data.
• Developed and evaluated 490 Scree-CNN classifier models, analyzing the impact of CNN parameters on performance.

Main Results:

• The optimal Scree-CNN model, with a learning rate of 0.01, mini-batch size of 64, and validation frequency of 50, achieved 100% accuracy, sensitivity, specificity, and precision.
• Performance indicators were compared against the WHO-recommended Enzyme-Linked Immunosorbent Assay (ELISA) for Dengue Fever diagnosis.

Conclusions:

• The Scree-CNN model demonstrates exceptional potential for accurate and noninvasive early detection of Dengue Fever using salivary NS1 SERS spectra.
• This approach offers a significant advancement over current diagnostic methods, paving the way for rapid point-of-care testing.