Diagnosis of odontogenic keratocysts and non-keratocysts using edge attention convolution neural network.

Background: The study's objective was to develop an automated method for a histopathology recognition model for odontogenic keratocysts (OKC) and non-keratocyst (Non-KC) in jaw cyst sections stained with hematoxylin (H) and eosin (E) on a tiny bit of incisional biopsy prior to surgery.

Methods: This hastens the speed and precision of diagnosis to patients. Also, navigates the clinicians with the therapeutic doctrine. To build such a system and to increase the accuracy of the existing models, the edge attention CNN model with Keras functional API was implemented which efficiently analyzes the texture information of the images. Approximately 2861 microscopic images at a 40X magnification were taken from 54 OKC, 23 Dentigerous cysts (DC), and 20 Radicular cysts.

Results: The model was trained using both RGB and canny edge-detected images. The model gave a good accuracy of 96.8%, which is suitable for real-time. Histopathological images are better analyzed through textural features. The proposed edge attention CNN highlights the edges, making texture analysis more precise.

Conclusions: The suggested method will work for OKC and Non-KC diagnosis automation systems. The use of a whole slide imaging scanner has the potential to increase accuracy and remove human bias.