Oral cancer diagnosis based on gated recurrent unit networks optimized by an improved version of Northern Goshawk optimization algorithm

Oral cancer early diagnosis is a critical task in the field of medical science, and one of the most necessary things is to develop sound and effective strategies for early detection. The current research investigates a new strategy to diagnose an oral cancer based upon combination of effective learning and medical imaging. The current research investigates a new strategy to diagnose an oral cancer using Gated Recurrent Unit (GRU) networks optimized by an improved model of the NGO (Northern Goshawk Optimization) algorithm. The proposed approach has several advantages over existing methods, including its ability to analyze large and complex datasets, its high accuracy, as well as its capacity to detect oral cancer at the very beginning stage. The improved NGO algorithm is utilized to improve the GRU network that helps to improve the performance of the network and increase the accuracy of the diagnosis. The paper describes the proposed approach and evaluates its performance using a dataset of oral cancer patients. The findings of the study demonstrate the efficiency of the suggested approach in accurately diagnosing oral cancer. Oral cancer early diagnosis is a critical task in the field of medical science, and one of the most necessary things is to develop sound and effective strategies for early detection. The current research investigates a new strategy to diagnose an oral cancer based upon combination of effective learning and medical imaging. The current research investigates a new strategy to diagnose an oral cancer using Gated Recurrent Unit (GRU) networks optimized by an improved model of the NGO (Northern Goshawk Optimization) algorithm. The proposed approach has several advantages over existing methods, including its ability to analyze large and complex datasets, its high accuracy, as well as its capacity to detect oral cancer at the very beginning stage. The improved NGO algorithm is utilized to improve the GRU network that helps to improve the performance of the network and increase the accuracy of the diagnosis. The paper describes the proposed approach and evaluates its performance using a dataset of oral cancer patients. The findings of the study demonstrate the efficiency of the suggested approach in accurately diagnosing oral cancer. Oral cancer is a kind of cancer which grows in throat or mouth tissues, which can be called cancer of oral cavity or mouth, and can affect areas such as the gums, floor or roof of the mouth, cheeks, tongue, lips, throat, and tonsils [1Cai X. et al.Breast cancer diagnosis by convolutional neural network and advanced thermal exchange optimization algorithm.Computational and Mathematical Methods in Medicine. 2021; 2021Crossref Scopus (99) Google Scholar]. Early diagnosis and cure, which has been considered crucial as cancer of oral, can be life-threatening [2Liu Haozhi Ghadimi Noradin Hybrid convolutional neural network and Flexible Dwarf Mongoose Optimization Algorithm for strong kidney stone diagnosis.Biomedical Signal Processing and Control. 2024; 91106024Crossref Scopus (6) Google Scholar]. Medical imaging techniques such as MRI (Magnetic Resonance Imaging), PET (Positron Emission Tomography), and CT (Computed Tomography) are vital to diagnose mouth cancer at the beginning stages [3Han Mengdi et al." Timely Detection of Skin Cancer: An AI-Based Approach on the basis of the Integration of Echo State Network and Adapted Seasons Optimization Algorithm".Biomedical Signal Processing and Control. 2024; 94 (nergy)Crossref Scopus (0) Google Scholar]. CT scans while making use of X-rays give detailed images of the oral cavity, including its blood vessels (arteries), soft tissues, and bones. The existence of oral cancer can be determined by CT scans by looking for abnormal growths or tumors in the mouth, throat, or neck [4Rajinikanth, V., N. Razmjooy, and N. Ghadimi, Design of a system for melanoma diagnosis using image processing and hybrid optimization techniques, in Frontiers of Artificial Intelligence in Medical Imaging. 2022, IOP Publishing.Google Scholar]. MRI scans utilize radio weaves and powerful magnetic fields to present some images, which are of the mouth cavity that reveal more specific details about the soft tissues [5Yu D. et al.Energy management of wind-PV-storage-grid based large electricity consumer using robust optimization technique.Journal of Energy Storage. 2020; 27101054Crossref PubMed Scopus (171) Google Scholar]. Throat or mouth tumors can be found with an MRI scan, along with any changes in the surrounding tissue that can point to the presence of malignancy [6Razmjooy N. et al.Computer-aided diagnosis of skin cancer: a review.Current Medical Imaging. 2020; 16: 781-793Crossref PubMed Scopus (92) Google Scholar]. PET scans track the body's metabolic activities by using a little quantity of radioactive materials. Because cancer cells have greater metabolic activity than healthy cells, PET scans can locate potential cancerous regions in the body even before a tumor develops [7Li Shunlei et al.Evaluating the efficiency of CCHP systems in Xinjiang Uygur Autonomous Region: an optimal strategy based on improved mother optimization algorithm.Case Studies in Thermal Engineering. 2024; 54104005Crossref Scopus (18) Google Scholar]. To confirm the presence of malignant cells in the oral cavity, clinicians can carry out a biopsy in addition to these imaging modalities. There is a practice called biopsy in which a tiny tissue sample is taken from the afflicted area and is inspected under a microscope [8Bo G. et al.Optimum structure of a combined wind/photovoltaic/fuel cell-based on amended Dragon Fly optimization algorithm: a case study.Energy Sources, Part A: Recovery, Utilization, and Environmental Effects. 2022; 44: 7109-7131Crossref Scopus (104) Google Scholar]. However, these methods can be time-consuming and invasive that can have restrictions with regard to effectiveness and the precision. Therefore, oral cancer diagnosis requires efficient and meticulous strategies [9França, R.P., et al., Using Metaheuristics in Discrete-Event Simulation, in Metaheuristics and Optimization in Computer and Electrical Engineering. 2020, Springer, Cham. p. 275-292.Google Scholar]. Recent developments in artificial Intelligence and machine learning demonstrate a promise for increasing the precision and effectiveness of cancer detection [10Zhang Li et al.A deep learning outline aimed at prompt skin cancer detection utilizing gated recurrent unit networks and improved orca predation algorithm.Biomedical Signal Processing and Control. 2024; 90105858Crossref PubMed Scopus (11) Google Scholar]. For instance, the objective of Ilhan et al. [11Ilhan B. et al.Improving oral cancer outcomes with imaging and artificial intelligence.Journal of dental research. 2020; 99: 241-248Crossref PubMed Scopus (0) Google Scholar] in 2020 was to assess the potential benefits of creating optical imaging techniques and some other strategies, which were based on AI for detecting mouth cancer and enhancing consequences. The paper gave a summary of clinical requirements and parameters of image-based approaches for identifying oral cancer. It also included a concise overview of research that employed these techniques separately or in combination, as well as information on artificial intelligence systems. While some cutting-edge imaging techniques have been researched to improve oral cancer outcomes, none had yet been widely adopted or had a major influence on practice or results. The study showed that merging artificial intelligence with imaging might significantly increase diagnostic accuracy, despite a lack of studies that focused on oral cancer. Recently, there were several works in terms of using the AI methods for this purpose [12Navid Razmjooy F.R.S. Ghadimi Noradin A Hybrid Neural Network – World Cup Optimization Algorithm for Melanoma Detection.Open Medicine. 2018; 13: 9-16Crossref Google Scholar]. Moreover, combining imaging and AI techniques might increase diagnosis precision, providing patients with oral cancer with improved options for therapy. Warin et al. [13Warin K. et al.Automatic classification and detection of oral cancer in photographic images using deep learning algorithms.Journal of Oral Pathology & Medicine. 2021; 50: 911-918Crossref PubMed Scopus (0) Google Scholar] in 2021 developed an automatic classification model to detect mouth cancer screening utilizing CNN (Convolutional Neural Network) that advanced learning algorithms. 700 oral photographs were used, 350 of which were oral squamous cell carcinoma and the rest were normal mouth mucosa. In order to classify and detect the illnesses, DenseNet21 and R-CNN were, in turn, utilized. The classification precision achieved a precision, recall, F1-score, sensitivity, specificity, and AUC of 99%, 100%, 99%, 98.75%, 100%, and 99%, respectively. The accuracy of detection achieved a precision, recall, F1-score, and AUC of 76.67%, 82.14%, 79.31%, and 0.79 respectively. In 2022, Panigrahi et al. [14Panigrahi S. Das J. Swarnkar T. Capsule network based analysis of histopathological images of oral squamous cell carcinoma.Journal of King Saud University-Computer and Information Sciences. 2022; 34: 4546-4553Crossref Scopus (27) Google Scholar] suggested a different method for the classification of mouth cancer using the capsule network, a deep learning technique. The capsule network was strengthened for rotation and affine transformation of the expanded oral information by dynamic routing and routing by agreement. The network's ability to handle orientation, position, and outlook made it appropriate for the early-stage analysis of mouth cancer images which were histopathological. The cross-validation performance indicated that the suggested strategy successfully classified the histopathological images of OSCC (Oral Squamous Cell Carcinoma) with sensitivity, specificity, and accuracy of 97.78%, 96.92%, and 97.35%, respectively. In 2023, Haron et al. [15Haron N. et al.High referral accuracy for oral cancers and oral potentially malignant disorders using telemedicine.Oral diseases. 2023; 29: 380-389Crossref Scopus (14) Google Scholar] assessed the effectiveness of a phone application called MeMoSA® to identify possible mouth cancers and malignant disorders. A prospective research was carried out with 355 participants; 280 of them were diagnosed with oral variants or lesions. MeMoSA® was contrasted to another oral test using kappa statistics based on some factors, like sensitivity, intra-rater agreement, concordance, F1-score, and specificity. The study found that specificity and sensitivity of MeMoSA® were, in turn, 95.5% and 94.0%, for referral decisions with inter-rater agreement of 0.825. An off-site expert reviewed the images; moreover, MeMoSA® documented regression or progression of the lesions systematically. In 2023, Gomes et al.[16Gomes R.F.T. et al.Use of artificial intelligence in the classification of elementary oral lesions from clinical images.International Journal of Environmental Research and Public Health. 2023; 20: 3894Crossref Scopus (10) Google Scholar] aimed to develop a CNN model to arrange six categories of oral images of lesion automatically. Four architectures were tested, and the InceptionV3 achieved the best results, whose correct predictions and average accuracy were more than 71% and 95.09%, respectively. The CNN model demonstrated satisfactory performance, which has potential for future use in identifying potentially malignant, malignant, and benign oral lesions. As can be observed from the literature, different models based on AI are proposed to the diagnose the oral cancer. Table 1 illustrates the key features of the literature.Table 1The key features of the literatureStudyKey FeaturesIlhan et al. (2020)Raman spectroscopy and autofluorescence imaging techniques were utilized. Both methods were combined for improving sensitivity and specificity.Warin et al. (2021)CNN-based models (DenseNet121 and ResNet-50) were combined for categorizing Oral Potentially Malignant Disorders (OPMDs) with high sensitivity and specificity.Panigrahi et al. (2022)A capsule network was proposed for oral squamous cell carcinoma classification, which was robust to rotation and affine transformations.Haron et al. (2023)The MeMoSA® mobile app was introduced for oral cancer identification using image review with high referral accuracy for early management.Gomes et al. (2023)Vision Transformer (ViT) and Swin Transformer were employed for mobile-based oral cancer image classification that could Achieve high accuracy rates. Open table in a new tab The present research showcases progress in artificial intelligence methods for identifying, categorizing, and forecasting oral cancer, which aids in prompt treatment and better results for patients. A new approach was explored for detecting oral cancer through the utilization of Gated Recurrent Unit (GRU) networks that have been fine-tuned by an enhanced version of the Northern Goshawk Optimization (NGO) algorithm. The benefits of this method encompass the ability to analyze extensive and intricate datasets, achieve high levels of accuracy, and identify cancer at an early stage. The upgraded NGO algorithm plays a crucial role in boosting the performance of GRU networks and improving diagnostic precision. One of the most widely-used and effective artificial intelligence tools is the Artificial Neural Network (ANN). Among them, Deep Neural Networks (DNN) have been observed to provide the highest efficiency and satisfactory results for various applications, including image and speech recognition. Sequential data can be processed by a kind of neural network called RNNs (Recurrent Neural Networks); moreover, RNNs have demonstrated promising findings in the medical field to diagnose and predict illness. Gated Recurrent Unit (GRU) networks, a kind of Recurrent neural network, can manage dependencies which are long-term and have also been used in the medical field for various applications. A metaheuristic optimization algorithm is a promising strategy for addressing optimization problems. One such algorithm is the Northern Goshawk Optimization (NGO) algorithm, which draws inspiration from the foraging behavior of Northern Goshawks. The NGO algorithm has demonstrated successful outcomes in various optimization problems and has shown potential in enhancing the performance of machine learning algorithms. Recently, numerous investigations have been conducted on the use of neural networks optimized by metaheuristic algorithms to diagnose oral cancer. These studies have demonstrated that the combination of networks and algorithms can enhance the precision and effectiveness to diagnose oral cancer. Nonetheless, myriad investigations are necessary to assess the potential of the current approach in clinical settings to distinguish its performance with other modern methods. The goal of the current essay is to propose a novel approach to diagnose oral cancer by utilizing GRU networks optimized by an enhanced version of the NGO algorithm. The suggested strategy intends to advance the precision and effectiveness of oral cancer diagnosis while also offering a non-intrusive and economical approach to diagnose oral cancer. Oral cancer is a significant public health issue worldwide, and if patients are timely detected and diagnosed, the chances of successful cure enhances. The utilization of ML (Machine Learning) with AI techniques, such as GRU networks optimized by an improved model of NGO algorithm, has demonstrated promise to enhance the precision and effectiveness of diagnosing oral cancer. The proposed approach in this research paper aims to provide a non-intrusive and economical strategy for diagnosing oral cancer; moreover, it has demonstrated high accuracy and specificity in diagnosing oral cancer. Further research can explore the clinical implementation of the proposed approach and validate its effectiveness in larger datasets. The paper provides a detailed description of the suggested method and the methodology used to evaluate its efficiency. The dataset used in the study consists of oral cancer patients, and the findings illustrate the efficiency of the suggested method in accurately diagnosing oral cancer. The study also highlights the potential of the proposed approach to improve the overall detection and treatment of oral cancer. The timely detection of oral cancer can significantly enhance the chances of successful treatment, and the proposed approach can help achieve this goal. In this research study, the "Oral Cancer (Lips and Tongue) images" information was utilized to analyze and evaluate the proposed method. The dataset consists of captured images from two distinct groups, comprising cancerous and non-cancerous, which can be employed to validate various types of diagnosis systems. The dataset comprises 87 sets of oral cancer images representing the cancerous group and 44 sets of oral non-cancer images representing the non-cancerous group. These images were captured by several Ear, Nose, and Throat (ENT) hospitals in Ahmedabad. To ensure accurate classification, the images were carefully examined and classified by ENT doctors, who provided their expertise in determining the presence or absence of oral cancer. All images in the dataset are saved in the widely-used "JPEG" format, denoted by the extension "*.jpg". Each image represents a specific case and offers visual information related to the presence or absence of oral cancer. The dataset provides a valuable resource for researchers, clinicians, and developers seeking to investigate and develop diagnostic systems of oral cancer. Access to the dataset is available through the Kaggle website, where it is hosted [17Shivam Barot, P.S. Oral Cancer (Lips and Tongue) images. 2020; Available from: https://www.kaggle.com/shivam17299/oral-cancer-lips-and-tongue-images.Google Scholar]. Researchers can retrieve the dataset from the specified URL and utilize it for various purposes, such as training machine learning models, developing computer-aided diagnosis systems, and conducting further studies in the context of detecting oral cancer. The "Oral Cancer (Lips and Tongue) images" dataset offers a valuable contribution to the research community and provides a foundation for advancing the understanding of oral cancer diagnosis and treatment. By using this dataset, researchers can explore novel approaches, algorithms, and techniques to upgrade the meticulousness and efficiency in detection of oral cancer, potentially causing timely diagnoses and enhanced patient consequences. Fig. (1(A, B)) displays the selection of sample images from the Oral Cancer Images (OCI) dataset, encompassing both cancerous and non-cancerous cases [17Shivam Barot, P.S. Oral Cancer (Lips and Tongue) images. 2020; Available from: https://www.kaggle.com/shivam17299/oral-cancer-lips-and-tongue-images.Google Scholar]. To confirm the diagnosis of oral cancer, a thorough physical examination of the patient's oral cavity is first conducted. Diagnosing oral cancer involves a combination of clinical examination, imaging tests, and tissue biopsy. While there is no official screening test for oral cancer, periodic oral self-exams are recommended to look for suspicious signs, such as abnormal lumps, sores, or color changes in the lips, gums, cheeks, and tongue. If anything is detected, an appointment should be scheduled promptly with an Ear-Nose-Throat (ENT) doctor. During routine healthcare visits, an ENT doctor, family physician, or dentist performs a head and neck examination, using a light and mouth mirror to visualize all surfaces inside the mouth and assess any abnormalities, such as sores, white patches (leukoplakia), or other changes. Here, the data has been separated into two folders, including "cancerous" and "non-cancerous". In this study, a method for oral cancer recognition is proposed based on JPEG. The methodology consists of four main stages, namely preprocessing, feeding the preprocessed data to the Gated Recurrent Unit (GRU) network, optimizing the network using the Northern Goshawk Optimization with Lévy flight and Chaos theory (NGO-LC), and final classification. In the initial stage of data preprocessing, images in JPEG format of oral cancer related to lip and tongue are imperiled to three techniques. These techniques include Random Over Sampling (ROS), data augmentation, and image resizing. The ROS technique is used to conduct the unbalanced data sets problem, where positive cases (oral cancer) meaningfully outnumber negative cases (non-cancer). Some techniques under the exclusive of data augmentation are employed to rise the size of a dataset unnaturally by performing a variety of transformations to the existing data. Moreover, an image resizing technique is used to ensure the standardization of image dimensions before entering the Gated Recurrent Unit network. The preprocessed data is subsequently used to train the GRU network. The GRU network is a kind of RNN that is particularly built to process sequential input, making it appropriate for processing oral images. The network undergoes training using preprocessed JPEG images to distinguish between malignant and non-cancerous instances. The Northern Goshawk Optimization method with Lévy flight and chaos theory is utilized to optimize the framework of the Gated Recurrent Unit network. The hunting habit of the Northern Goshawk, a superb bird of prey recognized for its quickness and agility in flight, inspired this algorithm. To achieve a balance between exploration and making use of the solution space, the algorithm incorporates the benefits of both chaos theory and Lévy flight. This lets the GRU network discover the ideal structure based on its weight, improving its efficacy in oral cancer identification. The Gated Recurrent Unit network is used for the final categorization of the oral pictures after it has been optimized using NGO-LC. It uses the learnt characteristics and weights to categorize the photos as malignant or non-cancerous. The study's suggested technique attempts to enhance oral cancer detection utilizing JPEG scans of the lips and tongue. The combination of preprocessing approaches, the GRU network and the NGO-LC optimization algorithm, is intended to improve classification accuracy and efficacy. Here, the pseudocode the suggested JPEG-based technique for the oral cancer identification has been illustrated using a GRU network with NGO-LC method:Algorithm 1Oral Cancer RecognitionInput: Oral cancer images in JPEG format1. Preprocessing: 1.1. Apply Random Over Sampling (ROS) to address imbalanced datasets. 1.2. Perform techniques of data augmentation to enhance size of the dataset. 1.3. Resize the images to a standardized dimension.2. Feed Preprocessed Data to GRU Network: 2.1. Train a Gated Recurrent Unit (GRU) network on the preprocessed data. 2.2. Input the preprocessed JPEG images into the GRU network. 2.3. Train the GRU network to distinguish between cancerous and non-cancerous cases.3. Optimize the Network Using NGO-LC: 3.1. Apply Northern Goshawk Optimization with Lévy flight and Chaos theory (NGO-LC). 3.2. Use the NGO-LC algorithm to optimize the structure of the GRU network based on its weights.4. Final Classification: 4.1. Utilize the optimized GRU network for the final classification. 4.2. Classify the oral images as malignant or non-cancerous based on the learned features and weights.Output: Classification results indicating whether the oral images are malignant or non-cancerous Open table in a new tab In the following, the different parts of this process have been explained. Image preprocessing methods, such as random oversampling (ROS), data augmentation, and image scaling, are critical in enhancing the efficacy of systems for diagnosing oral cancer images. Here a breakdown of each strategy has been revealed and how it helps improve system performance: By recognizing this fact that the dataset suffers from class imbalance, where the number of cancerous image cases is 87, while the number of non-cancerous image cases is 44. Class imbalance can lead to biased model training and lower performance in detecting the minority class, which in this case is the cancerous images. ROS is a method employed to handle imbalanced data sets in which the amount of positive cases (oral cancer) is much lower than the number of negative instances (non-cancerous). Imbalanced datasets might result in a biased model training and poor detection of the minority class. To balance the dataset, Random Over Sampling requires replicating instances from the minority class. By increasing the number of positive instances, the model obtains enough training on the oral cancer pictures, allowing for improved accuracy in classification and lowering the probability of misclassifying malignant cases as non-cancerous. Therefore, in this study, duplicating some of the non-cancerous images to increase their representation in the dataset is required. The oversampled non-cancerous images have been randomly selected from the original dataset until the number of cancerous and non-cancerous samples duplicate them so that they have become the same. Here, the number of cancerous image cases is 87, therefore, we need to add 43 (87-44) oversampling non-cancerous image cases to provide the same number of cases. The results in Fig. (2) demonstrate that how classes are distributed in a dataset prior to and subsequent to using an over-sampling strategy. The classifications under consideration include "non-cancerous" and "cancerous" images, which most likely correspond to either the existence or lack of oral cancer in JPEG photos. Data augmentation is applying different randomized alterations to existing data to artificially expand the dataset size. Augmentation of data helps the model generalize better to different variants and decreases the danger of being overfit by incorporating these variances. Data augmentation can imitate differences in image acquisition settings, such as varied lighting conditions, angles, or orientations of the mouth pictures, in the context of oral cancer imaging diagnosis. This allows the model to learn more robust characteristics that are more reflective of real-world settings, resulting in better performance in categorizing oral cancer images. In this study, Resizing, Warping, Cropping, Simulation noise, Simulation blur, Jitter color, and Translation have been used for images augmentation, which are explained in details in the following. Resizing involves adjusting the dimensions of an image while maintaining its aspect of ratio. It can be useful for standardizing the size of images in a dataset. In the present research, resizing has been likely applied to ensure that all oral cancer images have a consistent dimension before further processing. In this study, [0.5, 2, 3] resizing times are used for augmentation. Warping refers to geometric transformations that distort the shape of an image. The warping was applied between the range [40,60] degrees, possibly varying the amount of distortion. Warping can help introduce variations in images of oral cancer, making the model more robust to different shapes and deformations. Cropping involves selecting a specific region of interest from an image by removing unwanted portions. It helps focus the model's attention on the relevant features of the oral cancer images and discard irrelevant or distracting background information. Adding simulated noise to images introduces random variations in pixel values. Noise can simulate real-world imperfections in image acquisition and enhance the model's ability to be generalized to noisy or less pristine images. The level of noise added in this research is based on , with a higher value indicating more pronounced noise. Simulation blur involves introducing a level of blurriness to images. In this study, Gaussian blur with is utilized. This technique can mimic out-of-focus images, which can occur in real-world scenarios. By training the model on blurred images, it can become more robust to variations in image clarity and improve its generalization capabilities. Jittering color involves modifying the color attributes of the images. Here, some modifications have been conducted, namely a Contrast to 0.4, Hue to 0.1, Saturation to 0.2, and Brightness to 0.3. This technique helps the model learn invariant features across different color variations, making it more adaptable to changes in lighting conditions or color distributions. Translation refers to shifting an image along the horizontal or vertical axis. By randomly translating images within a specified range, the model can learn to recognize oral cancer features regardless of their position within the image. In this study, random transition between the range [-50,50] pixels has been assumed. By applying these data augmentation techniques, the aim was to enhance the performance of the oral cancer detection model. These techniques introduce variations, deformations, and imperfections to the images, allowing the model to learn from a more diverse and representative dataset. The purpose is to enhance the model's capability to be generalized; moreover, the aim was to accurately classify oral cancer images, even in the presence of variations in shape, noise, blur, color, and spatial positioning. It is important to note that the specific parameter values mentioned (e.g., sigma values) may vary depending on the research implementation and dataset characteristics. Fig. (3) depicts examples of original images that have undergone data augmentation techniques. Fig. (3) presents a series of examples that illustrate how data augmentation techniques can be used to create new images from an original image. The original image is show