Multi-Parameter Deep Learning Combined With Fluorescence Lifetime Imaging Microscopy for Non-Invasive and Label-Free...

Area of Science:

• Biomedical Optics
• Cancer Diagnostics
• Machine Learning in Medicine

Background:

• Endometrial cancer screening lacks noninvasive, accurate methods.
• Autofluorescence analysis offers potential for label-free cellular assessment.
• Metabolic changes in cancer cells can be detected via fluorescence lifetime imaging microscopy (FLIM).

Purpose of the Study:

• To develop and validate a noninvasive, label-free screening approach for endometrial cancer.
• To evaluate the efficacy of combining FLIM with multi-parameter deep learning for cancer risk prediction.
• To assess the diagnostic performance of metabolic parameters derived from NAD(P)H autofluorescence.

Main Methods:

• Collected cervical exfoliated cells from 71 participants.
• Utilized FLIM to detect NAD(P)H autofluorescence signals.
• Extracted mean fluorescence lifetime (t_m) and protein-bound fraction (a₂) as metabolic parameters.
• Developed single-parameter and multi-parameter deep learning models for prediction.

Main Results:

• The multi-parameter deep learning model integrating t_m and a₂ significantly outperformed single-parameter models.
• The multi-parameter model achieved 100% sensitivity and 92% specificity in external testing.
• Area Under the Curve (AUC) for the multi-parameter model was 0.92, an improvement of 0.17-0.31 over single-parameter models.

Conclusions:

• Combining multi-parameter deep learning with FLIM shows significant potential for endometrial cancer risk prediction.
• This approach offers a novel, noninvasive strategy for clinical cancer screening.
• The method leverages metabolic alterations in cells for accurate diagnostics.