Infrared Imaging Combined with Machine Learning for Detection of the (Pre)Invasive Pancreatic Neoplasia

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Cancer Diagnostics

Background:

• Pancreatic ductal adenocarcinoma (PDAC) has a low five-year survival rate (13%) due to limited early detection.
• Current pathological grading of pancreatic tissue is costly and time-consuming.
• Automated diagnostic systems are needed to revolutionize PDAC detection and treatment initiation.

Purpose of the Study:

• To develop an automated, stain-free diagnostic approach for pancreatic cancer detection.
• To combine Fourier-transform infrared (FT-IR) imaging with machine learning for PDAC analysis.
• To create models distinguishing normal, pre-malignant, and malignant pancreatic tissues.

Main Methods:

• Utilized stain-free Fourier-transform infrared (FT-IR) imaging combined with machine learning.
• Trained a Random Forest classifier on pancreatic tissues from a mouse model (KC/KPC) mirroring human PDAC progression.
• Analyzed tissues for normal structures, pancreatic intraepithelial neoplasia (PanIN), cancerous regions, hemorrhages, and collagen fibers.

Main Results:

• Developed a comprehensive model accurately distinguishing normal tissue from various pathological features (PanIN, cancer, hemorrhages, collagen).
• Created a streamlined model for rapid identification of normal versus pathologically altered tissues, including PanINs.
• Achieved accurate, cellular-level detection of PDAC indicators.

Conclusions:

• Stain-free FT-IR imaging and machine learning provide highly accurate diagnostic tools for early PDAC detection.
• This automated approach can significantly improve the chances for timely therapeutic intervention.
• Facilitates the shift towards automated tissue screening systems for improved cancer diagnostics.