Laryngeal Cancer Detection and Classification Using Deep Learning on Histopathological Images

Laryngeal cancer has complex and subtle symptoms, which pose challenges in its early diagnosis and detection. Our research applies deep-learning-based image classification models to classify and detect the early stages of laryngeal cancer using histopathological images. Through our study, a dataset contains a collection of 1320 tissue images augmented to 5280 images which are divided into Healthy (He), Hypertrophic Blood Vessels (Hbv), IPCL-like Vessel (IPCL), and Leukoplakia (Le), preprocessed and augmented to enhance and optimize the model performance. Several deep-learning and CNN-based pre-trained models like ResNet50, DenseNet121, DenseNet201, VGG16, VGG19, MobileNetV2, and InceptionV3, were tested. We have taken our dataset from an Italian researcher. Our fine-tuned Resnet50 model has achieved a high accuracy of 99.62%, which surpasses all the previous works. Along with the accuracy our goal in this research was to improve the precision and recall of the models as it is sensitive to disease classification.