London: Artificial intelligence (AI) may help doctors better predict the risk of patients developing oral cancer by ensuring accuracy, consistency and objectivity, according to researchers from the University of Sheffield in the UK.
The researchers are examining the use of AI and machine learning — the study of computer algorithms that improve automatically through experience — to assist pathologists and improve the early detection of oral cancer.
The rate of people being diagnosed with oral cancers including mouth, tongue, tonsil and oropharyngeal cancer, has increased by almost 60 per cent in the last 10 years, the researchers said in a statement.
Evidence suggests tobacco and alcohol consumption, viruses, old age as well as not eating enough fruit and vegetables can increase the risk of developing the disease, they said.
Oral cancer is often detected late which means that the patient survival rates are poor.
Currently, doctors must predict the likelihood of pre-cancerous changes, known as oral epithelial dysplasia (OED), developing into cancer by assessing a patient’s biopsy on 15 different criteria to establish a score.
This score then determines whether action is needed and what treatment pathway should be taken.
However, this score is subjective, which means there are often huge variations in how patients with similar biopsy results are treated.
For example, one patient may be advised to undergo surgery and intensive treatment, while another patient may be monitored for further changes.
“The precise grading of OED is a huge diagnostic challenge, even for experienced pathologists, as it is so subjective,” said Dr Ali Khurram, Senior Clinical Lecturer at the University of Sheffield’s School of Clinical Dentistry.
“At the moment, a biopsy may be graded differently by different pathologists. The same pathologist may even grade the same biopsy differently on a different day,” Khurram noted.
He said correct grading is vital in early oral cancer detection to inform treatment decisions, enabling a surgeon to determine whether a lesion should be monitored or surgically removed.
“Machine learning and AI can aid tissue diagnostics by removing subjectivity, using automation and quantification to guide diagnosis and treatment,” Khurram said.
“Until now this hasn’t been investigated, but AI has the potential to revolutionise oral cancer diagnosis and management by ensuring accuracy, consistency and objectivity,” he said.
Samples of archived OED tissue samples with at least five years of follow up data will be used in order to train AI algorithms and learn the statistical correlations between certain classifiers and survival rates.
These algorithms will aid pathologists in their assessment of biopsies helping them to make a more informed and unbiased decision about the grading of the cells and the patient’s treatment pathway.
The proposed algorithms have a strong translational angle and a potential to be rapidly deployed as an aid to clinical and diagnostic practice worldwide.
“People often feel threatened by AI, however rather than replacing a doctor’s expertise, exceptionally high-level of training and experience, the technology can help to assist their decision-making and compliment their skills,” said Khurram.
“This will help them to give a more accurate assessment and enable them to recommend the most beneficial treatment pathway for individual patients which we hope will help to improve survival rates,” he said.
According to Professor Nasir Rajpoot, from the University of Warwick in the UK, the pilot project will pave the way towards the development of a tool that can help identify pre-malignant changes in oral dysplasia, which is crucial for the early detection of oral cancer.
“Successful completion of this project carries significant potential for saving lives and improving patient healthcare provision,” said Rajpoot, one of the researchers in the study.
PTI