Objective:

To explore the use of AI applications in the diagnosis and risk assessment of oral epithelial dysplasia (OED) within the context of a Cancer Research UK-funded project.

Approach:

• Research Background: Adam Shephard discusses his transition from neuroimaging to computational pathology, focusing on OED as part of a Cancer Research UK-funded project.
• Challenges in Grading OED: The subjective nature of histological grading leads to variability among pathologists and does not reliably predict clinical outcomes.
• AI Model Development: HoVer-Net+ was developed to analyze histology images at the nuclear level, extracting biologically meaningful features for predicting cancer risk.
• Risk Prediction Importance: AI models aim to improve early detection and treatment by providing consistent risk assessments to support clinical decision-making.
• Biomarkers Identification: The study identified new biomarkers, such as increased lymphocyte infiltration, linked to malignant progression.
• Performance Comparison: AI-derived biomarkers showed strong agreement with expert pathologists, though performance varied with external datasets.
• Multimodal AI Models: Combining histology with clinical data offers a more comprehensive understanding of patient risk factors.

Key Findings:

• AI models can provide more consistent risk predictions than traditional histological grading.
• New biomarkers linked to cancer risk were identified, enhancing understanding of disease progression.
• AI-derived tools achieved performance comparable to expert pathologists in risk assessment.

Interpretation:

AI applications in OED diagnostics may enhance risk assessment and support clinical decision-making.

Limitations:

• Variability in performance when tested on external datasets due to differences in clinical practices.
• Challenges in collecting consistent clinical data across multiple sites for multimodal studies.

Conclusion:

AI has the potential to improve risk assessment in OED.