Deep Learning based Diagnosis of Jaw Cyst using Hematoxylin and Eosin Stained Incisional Biopsies Microscopic Images

Radicular cyst, Dentigerous cyst, and Odontogenic keratocyst should be differentiated due to their varying treatment approaches, recurrence rates, and biological behavior. Proper diagnosis ensures appropriate management, with Odontogenic keratocyst requiring more aggressive treatment and follow-up because of its higher recurrence and potential as-sociation with syndromes. Clinicians and pathologists may find it difficult to distinguish between a Radicular cyst, Dentigerous cyst, and Odontogenic keratocyst due to their complex micro-scopic diagnostic distinction. The development of computer-aided diagnostics for cyst identification leads to more accurate and timely diagnosis while reducing false positives and negatives. This work focused to design a deep learning based auto-mated jaw cyst diagnosis system that recognizes Radicular Cyst, Dentigerous Cyst and Odontogenic Keratocyst using Hematoxylin and eosin-stained images. Here, 27 Radicular cysts, 54 Odontogenic keratocysts, and 23 Dentigerous cysts patients's data were examined using a 40× magnification microscope. In this case, two stains were isolated, and convolutional layers were used to extract attributes from each stain image, Channel-wise attention was implemented to emphasize the important features. These attributes were concatenated, and relevant ones were selected by Relief F. A dense layer was used for classification. Obtained an accuracy of 95% on the test dataset