3D Hyperspectral Data Analysis with Spatially Aware Deep Learning for Diagnostic Applications.

Area of Science:

• Spectroscopy
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning (DL) is increasingly used in Raman spectroscopy data analysis.
• Current DL methods often ignore spatial information in 3D Raman hyperspectral scans.
• This limits the potential of analyzing complex tissue structures.

Purpose of the Study:

• To investigate the feasibility of using spatially aware deep learning algorithms for Raman spectroscopy.
• To enhance data analysis by preserving spatial information from 3D Raman hyperspectral scans.
• To improve the performance of deep learning models in tissue classification and cancer detection.

Main Methods:

• Applied a modified 3D U-Net for segmentation of Raman hyperspectral scans.
• Utilized a 3D convolutional neural network (CNN) for pixel-wise classification using Raman patches.
• Compared 3D spatially aware methods against a conventional 1D CNN baseline.
• Validated findings on colorectal and cholangiocarcinoma tissue datasets.

Main Results:

• Spatially aware deep learning models significantly increased performance in epithelial tissue and colorectal cancer detection.
• The 3D U-Net and 3D CNN approaches outperformed the 1D CNN baseline.
• Incorporating spatial information enhances model accuracy but may increase training complexity.

Conclusions:

• Preserving spatial information in 3D Raman hyperspectral scans is feasible and beneficial for deep learning.
• Spatially aware methods offer improved performance for spectroscopic data analysis, particularly in medical applications.
• Future research can leverage these findings for advanced spectroscopic data analysis tasks.