JOURNAL/crsat/04.03/02201859-202306020-00003/figure1/v/2023-08-03T140821Z/r/image-tiff
JOURNAL/crsat/04.03/02201859-202306020-00003/figure2/v/2023-08-03T140821Z/r/image-tiff
Background:
In patients with locally advanced head-and-neck squamous cell carcinoma (HNSCC), chemoradiotherapy improves outcomes. Radiotherapy commonly causes mucositis, which can significantly impede treatment and reduce the patient’s quality of life. Patients with severe mucocutaneous toxicity will show significant changes in thermal intensity early on, when compared to identically treated counterparts without toxicity.
Objective:
We aimed to assess the accuracy of the automated computer-aided deep learning approach in predicting the occurrence of oral mucositis in patients with HNSCC undergoing radiotherapy alone or with concurrent chemotherapy.
Materials and Methods:
This prospective observational study was conducted over four weeks in September 2021 in the Department of Radiotherapy at the Homi Bhabha Cancer Hospital, Sangrur (Punjab, India). We enrolled patients with HNSCC who were planned for radical intent radiotherapy, with or without concurrent chemotherapy. Using an automated deep learning technique, we analyzed the images taken with a FLIR-E60 thermal camera on the same day that patients received radiotherapy, with or without chemotherapy. Thermal images were binarily classified into two grades, that is, Grade 0 (absence of mucositis) and Grade I (asymptomatic or mild symptoms of mucositis). The dataset was split into training and testing cohorts, with a split ratio of 0.8. Accuracy was calculated as the ratio of correct predicted or classified instances to the total number of instances in the dataset. Accuracy was categorized as testing accuracy and training accuracy.
Results:
A total of 386 thermal images from 50 patients were acquired. Of these, 308 images (79.8%) were used for the training set and 78 (20.2%) for the testing set. There were 206 images (53.4%) with Grade 0 mucositis and 180 (46.6%) with Grade I. There was a significant difference in the thermal profile of patients with Grade 0 and Grade I images; P = 0.01. The model achieved promising results with 100% training accuracy and 82% testing accuracy. This led to a significant improvement in the false-negative rate of the proposed model, indicating improved performance.
Conclusion:
The deep learning approach-based analysis of thermal images can be a useful technique for predicting oral mucositis at an early stage in treatment, thus helping in intensifying supportive care. The model has been tested on a diverse dataset, and its performance in terms of accuracy validates the proposed model.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.