Applications of Data Characteristic AI-Assisted Raman Spectroscopy in Pathological Classification

Area of Science:

• Biomedical Spectroscopy
• Artificial Intelligence in Diagnostics
• Computational Pathology

Background:

• Raman spectroscopy enables label-free biomolecular analysis for pathological diagnosis.
• Artificial intelligence (AI) models like machine learning and deep learning enhance Raman spectroscopy-based disease diagnosis.
• Optimal AI model selection for diverse Raman spectral data characteristics remains a challenge.

Purpose of the Study:

• To explore the performance of various AI classification models on diverse Raman spectral datasets.
• To develop a data-characteristic-assisted AI classification model for optimizing AI performance.
• To improve diagnostic accuracy for conditions including cancer, bacterial infections, and diabetic skin complications.

Main Methods:

• Selected five representative Raman spectral datasets (endometrial carcinoma, hepatoma EVs, bacteria, melanoma, diabetic skin) with varying characteristics.
• Evaluated AI models including PCA-SVM, SVM, UMAP-SVM, ResNet, and AlexNet.
• Developed a data-characteristic-assisted AI model, optimizing parameters based on data size and KL divergence.

Main Results:

• Deep learning model ResNet outperformed PCA-SVM and UMAP on large spectral data size datasets.
• The data-characteristic-assisted AI model significantly improved accuracy across all tested datasets.
• Accuracy gains ranged from 53.7% to 85.5% for diabetic skin screening and 89.3% to 99.7% for melanoma cell detection, with a mean time expense of 5 seconds.

Conclusions:

• A data-characteristic-assisted AI approach effectively optimizes AI model selection for Raman spectroscopy.
• This method enhances diagnostic accuracy in various pathological conditions.
• The optimized AI models offer efficient and accurate label-free biomolecular analysis for clinical applications.