Abstract Deep learning – Convolutional Neural Networks (DL-CNN) has shown a lot of potential in identifying cancerous and non-cancerous oral lesions from oral photographic images. Moreover, the accuracy of CNNs can be improved by guiding the model to concentrate on the cancerous areas rather than not so important surrounding areas. The paper proposes to develop a DL-CNN model that directly focusses on cancerous areas in the lip, tongue and cheek images. The proposed CNN model applies transfer learning using DenseNet201 as base model for the detection. The model works by identifying region of interests (RoIs) and generating attention maps for the images that helps the model in understanding the area of focus by highlighting it and for classifying the oral lesions correctly. The results demonstrated effectiveness of the approach by giving an accuracy of 84.7% for the classification of oral cancer lesions.
Field : Mühendislik
Journal Type : Uluslararası
Relevant Articles | Author | # |
---|
Article | Author | # |
---|