In Vivo Time-Resolved Fluorescence Detection of Liver Cancer Supported by Machine Learning.

Area of Science:

• Biomedical optics
• Medical physics
• Cancer diagnostics

Background:

• Time-resolved fluorescence (TRF) is a key optical biopsy method for assessing cellular metabolism and antioxidant defenses.
• Cancerous cells exhibit distinct metabolic shifts, differing from healthy liver tissue.
• Machine learning (ML) excels at classification tasks in medical practice, including biomedical optics.

Purpose of the Study:

• To integrate TRF measurements with ML algorithms for automated classification of liver tissue.
• To differentiate between healthy liver parenchyma and various liver tumor types (primary malignant, metastases, benign).

Main Methods:

• Optical biopsies were conducted using a fine-needle optical probe under ultrasound guidance.
• Fluorescence decays were recorded from liver lesions and healthy tissue during percutaneous biopsy.
• ML models, including random forest, were trained and evaluated on histopathologically classified data.

Main Results:

• TRF spectroscopy revealed unique metabolic shifts characteristic of different liver tumor types.
• The ML approach reliably separated liver tissue into cancer and non-cancer classes with high accuracy (>0.90).
• The method achieved preliminary diagnosis of liver tumor types with sensitivity, specificity, and accuracy exceeding 0.80, 0.95, and 0.90, respectively.

Conclusions:

• This combined TRF and ML approach demonstrates significant potential for liver cancer diagnosis.
• The findings support its role as a valuable tool for developing future diagnostic and therapeutic strategies for liver cancers.