Automated Oral Epithelial Dysplasia Grading Using Neural Networks and Feature Analysis
Neda Azarmehr, Adam Shephard, Hanya Mahmood, Nasir Rajpoot, Syed Ali Khurram
Oral epithelial dysplasia (OED) is a precancerous lesion, histologically graded as mild, moderate or severe. The manual histological diagnosis of OED is time-consuming and subjective. We explore a customised Neural Architecture Search (NAS) technique to optimise an efficient architecture for full epithelium and individual nuclei segmentation in pathology whole slide images (WSIs). Results show the NAS-derived model outperforms all state-of-the-art networks. Accurate nuclear segmentation allows us to extract morphometric features. We propose a random forest model, using these features, to differentiate between OED grades.
Thursday 7th July
Poster Session 2.2 - onsite 11:00 - 12:00, virtual 15:20 - 16:20 (UTC+2)