The use of multilayer perceptron and radial basis function: an artificial intelligence model to predict progression of oral cancer
Nivedita Jayaram, Muralidharan Manjusha, Muthupandian Saravanan · International Journal of Surgery · 2023
Dear Editor, The progression of oral cancer (OC) is usually viewed as an ongoing series of epithelial changes in the oral mucosa1. Due to its advanced locoregional stage at the time of diagnosis, OC ranks as the sixth most often reported malignant disease worldwide, with significant rates of cancer morbidity and mortality. The greatest outcomes for patients come from early discovery and fast treatment, although the preponderance of OC tumours is found in late stages, with a 55% mortality rate. Poor OC results are mostly caused by unavailable or insufficient local screening and surveillance, which leads to delays in expert referrals and possible treatments2. Potential therapeutic conundrums could be alleviated by comprehending the advancements of innovations like artificial intelligence (AI). The use of AI in the treatment of oral malignant tumours can help with the current woes in prognosis prediction and illness diagnostics. AI imitates human cognitive processes, represents an innovation breakthrough and has captured the attention of scientists all around the world. AI system is a framework that analyses information to find designs, trains itself using data and produces outcomes effective outcomes3 (Fig. 1).Figure 1: Workflow of an artificial intelligence model.A pathologist can assess whether a patient has malignancy and the stage of the disease when histopathologic samples are tested for OC. There is occasionally a risk for error because human quantification is necessary for the evaluation of samples for diagnostic features with error invariably resulting in erroneous altercation. As a result, AI has decreased these mistakes and enhanced the effectiveness and precision of recognizing the cytologic and histologic aspects of OC. AI technology can also analyse big sample sizes to find OC. In the studies that were chosen, biopsy and histologic samples as well as photographic pictures were used as samples. Biopsy and histologic samples were used in six research. In several investigations, cellular alterations were used as a marker to distinguish malignant samples from regular and aberrant cell nuclei. Using the suggested segmentation method, examined epithelial alterations by finding keratin pearls in the oral mucosa of patients with OC. With their suggested convolutional neural network (CNN) machine, they quantified the keratinization layer, which was successful because this metric is crucial for diagnosing the stage of OC. For the purpose of detecting OC, we have several deep learning algorithms available. CNN is one such method that might be the most effective at finding OC. A more sophisticated and promising variation of the traditional artificial neural network model is CNN. It is designed to handle more complicated problems, preprocessing and data compilation. It uses the order in which the neurons are arranged in its datasets as a point of reference. The convolutional layer, the pooling layer, the ReLU correction layer and a fully connected layer similar to a multilayer perceptron are the four different types of layers used in this method. Based on the number of outputs, we can estimate the number of neurons, the complexity and whether it is benign or cancerous when you run an magnetic resonance imaging scan over these layers4,5. Radial basis function neural network (RBF) is yet another method to understand the data. An artificial neural network that uses radial basis functions as activation functions is known as an RBF network. A linear combination of the inputs’ radial basis functions and the neuronal parameters makes up the network’s output. These kinds of functions are used for both classification and regression and can take several inputs at once. RBF employs weights and a variety of RBF functions, and we can change the weights and RBF parameters to suit each given case. The Gaussian or normal distribution is the most straightforward to implement among the several functions used to estimate data distribution, as you only need to figure out the mean and standard deviation for the training data (Fig. 2). A large number of additional species may be added, but only related solution vectors may cross over. This module increases the accuracy rate by intensifying the nucleus pixels6,7.Figure 2: Schematic representation of simple multilayer perceptron, radial basis function and convolutional neural network.AI will significantly alter studies on the early diagnosis of OC, and so enhance medical treatment in general. Though AI is in its primitive stage, diagnoses of patients, clinical judgement calls and failure predictions in the dental field can all be successfully performed using AI. It is a trustworthy modality for use in the fields of OC detection along with application in other fields such as periodontics, prosthodontics, orthodontics, forensic dentistry, radiography and oral and maxillofacial surgery as well as restorative dentistry8. AI provides significant prospects for task automation by detecting complicated patterns. In this regard, research is critical to facilitating the multidisciplinary adoption of such methodologies, and advancements in this area could pave the way for future investigations9. Early identification is crucial for patients with OC because the disease’s prognosis is dismal in its late stages. Using the information from cytology images, fluorescence images, computed tomography images and depth of invasion, OC can be diagnosed more rapidly and accurately. On the basis of the risk classification model, 11 981 prepossessed pictures were loaded for AI analysis to evaluate the effectiveness of AI with traditional cytology and histology and the results indicated a diagnostic accuracy of 80–84%3. Machine learning produces reliable outcomes for identifying OC, which is beneficial for pathologists to enhance their diagnostic outcomes and reduce the possibility of error. Furthermore, deep learning (neural networks), which denotes superior efficiency and hence is more accurate, has been employed in the research that were considered the best based on the data. It is evident that when compared with traditional methods of diagnosis, AI is more accurate in diagnosing OC. Using reliable data from AI, patients can be categorized as high risk or low risk, assisting physicians in diagnosis and treatment of OC. Ethical approval None. Sources of funding None. Authors’ contribution N.J.: investigation, writing – original draft preparation. M.M.: data collection, visualization, writing – original draft preparation. S.M.: conceptualization, writing – reviewing and editing, visualization and supervision. Conflicts of interest disclosure The authors declare that they have no financial conflict of interest with regard to the content of this report. Research registration unique identifying number (UIN) None. Guarantor Saravanan Muthupandian. Provenance and peer review Not commissioned, internally peer-reviewed. Data statement All data are available in the manuscript.