Deep Learning Fuzzy Inference: An Interpretable Model for Detecting Indirect Immunofluorescence Patterns Associated...

Area of Science:

• Oncology
• Artificial Intelligence
• Immunology

Background:

• Nasopharyngeal cancer (NPC) screening relies on Epstein-Barr virus (EBV) antibody detection via immunofluorescence assay (IFA).
• Early NPC detection significantly improves survival rates, highlighting the need for scalable screening methods.
• Current IFA methods are labor-intensive due to manual interpretation, limiting their scalability.

Purpose of the Study:

• To develop and validate an automated system for NPC detection using AI.
• To improve the scalability and accuracy of NPC screening compared to traditional IFA methods.

Main Methods:

• Development of an automated fuzzy inference (FI) system for analyzing cellular staining patterns.
• Integration of a deep learning module into the FI system, creating a hybrid model named deep learning FI (DeLFI).
• Clinical validation of the DeLFI model using a separate testing set of patient samples.

Main Results:

• The initial FI system showed high agreement with human experts (κ = 0.82).
• The hybrid DeLFI model achieved improved performance (κ = 0.90) and reduced manual review cases.
• In clinical validation, DeLFI surpassed human evaluation in AUC (0.926 vs 0.821) and matched human performance in Youden J index (0.81 vs 0.80).

Conclusions:

• The DeLFI model demonstrates potential for enhancing the scalability and accuracy of NPC detection.
• Combining deep learning with fuzzy inference offers a promising approach for automated cancer screening.
• This AI-driven method could significantly impact early diagnosis and patient outcomes for NPC.