AI-Driven Large-scale Screening for Oral and Oropharyngeal Cancer
Time period: 2021-01-01 to 2023-12-31
Project leader: Joakim Lindblad
Type of award: Project grant
Total fundning: 2 500 000 SEK
Purpose and goal: High mortality from cancer of the oral cavity and throat is largely due to late diagnosis. Today´s methods for cancer detection are expensive and painful. Modern AI-techniques make it possible to radically reduce the cost of analysis while increasing accuracy. The project aims to scale up our developed deep learning-based methods for early cancer detection to reliable large-scale usage in healthcare in Sweden and India, with the goal of delivering cheap and effective AI-driven screening for oral and oropharyngeal cancer via painless brush samples and AI-based imaging diagnostics. Expected results and effects: The project enables reliable detection of patients at risk of developing cancer of the oral cavity and throat through non-invasive brush sampling. Cytological analysis is enhanced by an XAI-based diagnostic support system that reduces the burden of manual analysis and provides fast and reliable diagnosis at low cost. The project results, in the form of an economically efficient and practically simple, yet powerful screening program for early detection of oral and oropharyngeal cancer, will have a strong positive impact on global health, but also on the usefulness of AI in healthcare. Approach and implementation: Data collection and slide digitization. Evaluation of different sampling models. Use and further development of our CytoBrowser tool for interactive annotation and assessment, delivering training data for the AI learning steps. Development of AI-controlled microscopy for quality-assured rapid imaging. Further improving deep learning-based cytological analysis for reliable, well-differentiated early diagnosis. XAI-based results visualization in CytoBrowser for effective assessment and feedback for active few-shot learning and further improved performance.