Importance of complementary data to histopathological image analysis of oral leukoplakia and carcinoma using deep neural networks

• DOI: 10.1016/j.imed.2023.01.004
• Publication Date: 2023-02-06T18:59:27Z
• Authors: Leandro Muniz de ...; Maria Clara Falcã...; Júlia Pessini Soares; Tânia Regina Grão...; Liliana Aparecida...; Danielle Resende ...; Renato Antonio Kr...
• Keywords: Oral leukoplakia

Background Oral cancer is one of the most common types in men causing mortality if not diagnosed early. In recent years, computer-aided diagnosis (CAD) using artificial intelligence techniques, particular, deep neural networks have been investigated and several approaches proposed to deal with automated detection various pathologies digital images. Recent studies indicate that fusion images patient's clinical information important for final diagnosis. As such dataset does yet exist oral cancer, as far authors are aware, a new was collected consisting histopathological images, demographic data. This study evaluates importance complementary data image analysis leukoplakia carcinoma CAD. Methods A (NDB-UFES) from 2011 2021 information. The 237 samples were curated analyzed by pathologists generating gold standard classification. State-of-the-art architectures (Concatenation, Mutual Attention, MetaBlock MetaNet) latest learning backbones 4 distinct tasks identify squamous cell carcinoma, dysplasia without dysplasia. We evaluate them balanced accuracy, precision, recall area under ROC curve metrics. Results Experimental results best models present accuracy 83.24% ResNetV2 backbone. It represents an improvement performance 30.68% (19.54 pp) task differentiate or Conclusion confirms cured positively influence classification cancer.