Deep Learning-Based Classification of Laryngeal Abnormalities Using DenseNet Architectures

Laryngeal cancer is a malignant disease that develops in the voice box, which is located in the throat. Current early detection and diagnosis tools and techniques of laryngeal cancer meet a variety of performance bottlenecks, including decreased accuracy in early detection, high time complexity for computations, and longer screening procedures for patients. To overcome these limitations, the performance of three deep learning architectures DenseNet-121, DenseNet-169, and DenseNet-201 is evaluated. All these belong to the DenseNet family that is well regarded for their feature reuse in densely connected convolutionallayers. The evaluation focuses on overall performance, including metrics such as accuracy, precision, recall, F1-score, and computational efficiency, to determine the most effective model for early-stage laryngeal cancer classification. DenseNet gave improved and practical results and obtained an accuracy of 0.98, precision of 0.97, specificity of 0.99, F1 score of 0.97, and recall of 0.97.