Chronological spotted hyena optimized deep learning approach for laryngeal cancer detection using image and voice analysis

Laryngeal cancer leads to significant problems with swallowing, voice production, and overall quality of daily life. This research proposes a Chronological Spotted Hyena Optimizer-based Deep Residual Network (CSHO-based DRN) for the detection of laryngeal cancer. Initially, the input images and voice samples are acquired and pre-processed. Meanwhile, the extraction of image features like Global Binary Pattern (GBP), Histogram of Oriented Gradients (HoG) and multi-kernel Spider Local Image Features (SLIF) is performed from the preprocessed image samples. From the preprocessed speech sample, signal features like Bark Frequency Cepstral Coefficients (BFCC), Multiple Kernel Weighted Mel Frequency Cepstral Coefficients (MKMFCC), and spectral features, namely spectral skewness, spectral centroid, spectral spread, tonal and power ratio, are acquired. Finally, CSHO-based DRN is used to detect laryngeal cancer using the extracted features. The hyperparameters of DRN are optimized using the CSHO algorithm. The proposed CSHO algorithm merges the chronological concept with the Spotted Hyena Optimizer (SHO). The CSHO-based DRN attained 0.929 accuracy, 0.933 sensitivity and 0.961 specificity.