Radiotherapy is the main treatment strategy for nasopharyngeal carcinoma. A major factor affecting radiotherapy outcome is the accuracy of target delineation. Target delineation is time-consuming, and the results can vary depending on the experience of the oncologist. Using deep learning methods to automate target delineation may increase its efficiency. We used a modified deep learning model called U-Net to automatically segment and delineate tumor targets in patients with nasopharyngeal carcinoma. Patients were randomly divided into a training set (302 patients), validation set (100 patients), and test set (100 patients). The U-Net model was trained using labeled computed tomography images from the training set. The U-Net was able to delineate nasopharyngeal carcinoma tumors with an overall dice similarity coefficient of 65.86% for lymph nodes and 74.00% for primary tumor, with respective Hausdorff distances of 32.10 and 12.85 mm. Delineation accuracy decreased with increasing cancer stage. Automatic delineation took approximately 2.6 hours, compared to 3 hours, using an entirely manual procedure. Deep learning models can therefore improve accuracy, consistency, and efficiency of target delineation in T stage, but additional physician input may be required for lymph nodes.

No abstract

No abstract

We acknowledge Dr. Wen Liao for his valuable suggestions. *Jialing Liu and Ye Chen contributed equally to this work.

Oral squamous cell carcinoma (OSCC) is prevalent around the world and is associated with poor prognosis. OSCC is typically diagnosed from tissue biopsy sections by pathologists who rely on their empirical experience. Deep learning models may improve the accuracy and speed of image classification, thus reducing human error and workload. Here we developed a custom-made deep learning model to assist pathologists in detecting OSCC from histopathology images. We collected and analyzed a total of 2,025 images, among which 1,925 images were included in the training set and 100 images were included in the testing set. Our model was able to automatically evaluate these images and arrive at a diagnosis with a sensitivity of 0.98, specificity of 0.92, positive predictive value of 0.924, negative predictive value of 0.978, and F1 score of 0.951. Using a subset of 100 images, we examined whether our model could improve the diagnostic performance of junior and senior pathologists. We found that junior pathologists were able to delineate OSCC in these images 6.26 min faster when assisted by the model than when working alone. When the clinicians were assisted by the model, their average F1 score improved from 0.9221 to 0.9566 in the case of junior pathologists and from 0.9361 to 0.9463 in the case of senior pathologists. Our findings indicate that deep learning can improve the accuracy and speed of OSCC diagnosis from histopathology images.