Artificial Intelligence based assessment of oral precancer to aid early detection of oral cancer

Background Oral squamous cell carcinoma (OSCC) is amongst the top ten leading cancers worldwide, with an increasing incidence and worsening prognosis. UK OSCC rates have increased by 64% over the last decade, with almost 8,000 people affected each year. Despite advancement in medical and surgical techniques, less than 50% of OSCC patients survive longer than five-years.

Patients with OSCC are more likely to die than those with breast, lung, colorectal and cervical cancers emphasising the importance of early and accurate diagnosis. Oral Epithelial Dysplasia (OED) represents precancerous oral changes and is the precursor to OSCC. Around 30-80% of OED lesions can progress to malignancy, but with early intervention up to 70% can be prevented.

Histopathological analysis remains the diagnostic gold standard. However, oral precancer/OED grading based on the World Health Organisation criteria remains highly subjective.

Since the grade is used to determine cancer risk and treatment decisions, an inaccurate diagnosis can be detrimental to patients.

Artificial Intelligence (AI) can eliminate this variability by introducing objective grading and reproducibility, which can play a key role in early detection and improved patient management.

Aim To develop novel AI software to aid early and accurate oral precancer diagnosis and prediction of malignant transformation. Research Questions 1. Can AI accurately and consistently diagnose and grade oral precancer? 2. Can AI reveal novel information about the progression from oral precancer to cancer? 3.

Can AI be used to predict the risk of oral cancer development?

Methods High-resolution digital images of oral precancerous tissue, in addition to normal and cancerous tissue will be used to develop AI algorithms.

Cases will be identified from the pathology archive in Sheffield between 2008-2013 to yield five-year clinical follow-up data. Ethical approval is already in place.

Stage one will involve training the AI algorithms to detect specific histological features of OED (as outlined in WHO criteria). Stage two will use existing AI algorithms to identify novel features predictive of OED progression into cancer.

Stage three will develop a cancer-risk prediction tool correlating clinical information with significant features identified in the first two stages.

Potential Impact This research will be innovative and translational and has the potential to positively impact on the reduction of OSCC rates by aiding early and accurate precancer detection.

The AI software will provide patients and clinicians with faster and more reliable information about the diagnosis and prognosis of oral precancer, which will be crucial to guide future patient management.

Patients will be engaged throughout the study to obtain feedback on the ongoing work and results to enhance translational impact. Study findings will be widely shared with academic and clinical colleagues.

This study will be important to facilitate future multi-centric research with other UK hospitals and cancer centres internationally (including low-and middle-income countries) to further optimise the developed software and determine its impact on early and accurate detection.

It is anticipated that this study will lead to a clinical trial to provide patients and clinicians with objective information about the diagnosis and prognosis of suspicious oral lesions.