|Year : 2022 | Volume
| Issue : 2 | Page : 154-160
Application of artificial intelligence in the diagnosis and survival prediction of patients with oral cancer: A systematic review
S Canty Sandra1, Anusha Raghavan1, PD Madan Kumar1, 2
1 Department of Public Health Dentistry, Ragas Dental College and Hospital, Chennai, Tamil Nadu, India
2 , India
|Date of Submission||22-Sep-2021|
|Date of Decision||13-Mar-2022|
|Date of Acceptance||21-Mar-2022|
|Date of Web Publication||01-Jul-2022|
Department of Public Health Dentistry, Ragas Dental College and Hospital, 2/102, East Coast Road, Uthandi, Chennai - 600 119, Tamil Nadu
Source of Support: None, Conflict of Interest: None
Oral cancer constitutes around 2.1% and it is the sixth-most common malignancy worldwide and the third-most common type of malignancy in India. The purpose of this systematic review is to find the prediction of survival rate among oral cancer patients using artificial intelligence (AI) and its forms like machine learning. Suitable articles were identified by searching PubMed, Trip database, Cochrane, and Google Scholar host databases. The search was done with the help of PIO analysis where the population stands for oral cancer patients, the intervention given here were AI and its subsets and the outcome were diagnosis and survival prediction of oral cancer. The screening of the titles and abstracts was done, and only those articles that fulfilled the eligibility criteria were selected. The search resulted in 451 articles, of which only six articles that fulfilled the criteria were included. The studies showed that AI models were able to predict the 5-year survival rate among oral cancer patients. The accuracy of the decision tree classifier, logistic regression, and boosted decision tree models were 76%, 60%, and 88.7%, respectively. Modern age diagnosed people tend to have a longer survival rate than those diagnosed in the past. The limitation was that these studies were created using retrospective cohorts, but for validation, they must be compared with prospective cohorts. These studies are important for identification and survival prediction, which will contribute to future advancements, change in the treatment plan, and reduce health-care problems.
Keywords: Artificial intelligence, machine learning, mouth neoplasm, oral cancer, survival rate
|How to cite this article:|
Sandra S C, Raghavan A, Madan Kumar P D. Application of artificial intelligence in the diagnosis and survival prediction of patients with oral cancer: A systematic review. J Oral Res Rev 2022;14:154-60
|How to cite this URL:|
Sandra S C, Raghavan A, Madan Kumar P D. Application of artificial intelligence in the diagnosis and survival prediction of patients with oral cancer: A systematic review. J Oral Res Rev [serial online] 2022 [cited 2022 Dec 3];14:154-60. Available from: https://www.jorr.org/text.asp?2022/14/2/154/349713
| Introduction|| |
Oral cancer constitutes around 2.1% of all cancer. Early diagnosis increases the patient's survival rate. Around 60% of the mouth neoplasm patients survive up to 5 years. When diagnosed at an early stage, the survival rate is around 82%, while at later stages, the survival rate is only 27%. Decision-making tools based on artificial intelligence (AI) have shown to be helpful for the prediction of oral cancer survival rate. AI is the science of making machines to mimic the behavior of humans. AI has become our part of life and it is being used in our day-to-day activities in the form of recognizing our speech (Google assistant), voice assistant (Alexa, Siri), face recognition, and so on. Machine learning is an AI subfield that focuses on feeding data to machines (computers) to assist them in making decisions. Hence, this systematic review was done to find out the prediction of survival rate among oral cancer patients, which is done by AI.
| Materials and Methods|| |
To locate, analyze, and summarize all relevant study findings, the systematic review was conducted utilizing objective and transparent procedures in accordance with the preferred reporting items for systematic reviews and meta-analysis (PRISMA) Guidelines. For this systematic review, the protocol was originally registered with PROSPERO (International prospective registration of systematic reviews), ID: CRD42021239964.
The following is the Population, Intervention, Outcome (PIO) analysis of the articles found.
Patients reporting or undergoing treatment for oral cancer.
Various subsets of AI in the form of machine learning, neural networks, or deep learning.
Diagnosis and survival prediction of oral cancer, thereby improving the quality of life for patients.
Studies in which any form of machine learning, neural networking, or deep learning technology has been used for the diagnosis and prediction of survival of oral cancer patients.
- Studies from the past 8 years (2013–2021), which included cross-sectional, case-control, cohort, randomized, or nonrandomized observational or interventional study designs, were considered
- All those studies assessing any type of oral cancer according to their tumor, node, and metastasis (TNM) staging from T1 to T IVB with or without metastasis and nodal involvement were included
- The publications considered were those which were published only in the English language.
- Articles of the following type were excluded: Review articles, expert opinion, conference papers, blog posts, discussion articles, systematic reviews, and meta-analysis
- Publications without an abstract and those that were outside of the study's objective were excluded
- Studies with T0 and Tis tumor stages were excluded.
From January 2013 to January 2021, a broad literature search was conducted in PubMed, Trip, Cochrane, and the Google Scholar host database. The keywords used in the search were AI, machine learning, oral cancer, neural network, diagnosis, survival prediction. The combination of the following terms oral cancer and AI, machine learning, neural network and diagnosis, survival prediction were included in the search strategy. In PubMed, the articles were searched through the medical subject headings terminologies AI and machine learning. A hand search of articles was also conducted to ensure additional relevant references, but there were no relevant articles found. To find the relevant publications, all the full-text articles were searched [Table 1].
By reviewing all the titles and abstracts according to the exclusion and inclusion criteria, two authors (CSS and AR) individually selected papers published. A comparison of papers was completed between the two authors and there were no disagreements regarding inclusion. Using SPSS software (IBM corp. Statistical Package for Social Sciences software for windows; version 20.0 Armonk, New York), the inter-rater agreement between the authors was 0.87, which is a good result.
The data extraction from the final six articles was done using a data extraction form. It includes the first author's name, year of publication of the article, the aim of the study, objectives of the study, study design, study summary, results, and outcome [Table 2].
|Table 2: Data extraction of the final articles, including the first author name, year of publication of the article, the aim of the study, objectives of the study, study design, study summary, results, and outcome|
Click here to view
Quality assessment of the included studies
The final analysis included six cohort studies. The methodological quality of the selected articles was assessed using the Newcastle-Ottawa Quality Assessment Scale for Cohort Studies Wells et al. [Table 3]. For cohort studies, the quality score was based on three items of the following categories: Group selection, comparability, and outcome. A maximum of one star was given for the selection and outcome categories and a maximum of two stars was given for comparability. The higher the score under these categories, the better the quality of the study. It was rated as “good” quality if there were 3 or 4 stars in the selection category; 1 or 2 stars in the comparability domain and 2 or 3 stars in the outcome/exposure category. It was rated as “fair” quality if there were 2 stars in the selection category; 1 or 2 stars in the comparability category and 2 or 3 stars in the outcome/exposure category. It was rated as “poor” quality if there were 0 or 1 star in the selection category; 0 stars in the comparability category and 0 or 1 stars in the outcome/exposure category. If there were more than 7 stars it is rated as good quality. If there were 5 or 6 stars, it is rated as fair quality. If there were <5 stars, it is rated as poor quality. The results of the assessment are tabulated in [Table 3].
|Table 3: Quality assessment - Newcastle-Ottawa Quality Assessment Scale for Cohort Studies|
Click here to view
| Results|| |
A total of 451 articles were generated in the initial search. Finally, based on the inclusion and exclusion criteria, only six articles were selected for the review using the PRISMA flowchart [Figure 1]. These six articles were assessed with the help of this scale and the final results are three articles were rated as good quality and three articles were rated as fair quality. The authors CSS, AR rated the individual studies independently, which was further verified by the author, PDMK. The discrepancies between the first two authors were resolved by the consensus from the third author. Good inter-rater reliability (kappa value = 0.8) was present.
|Figure 1: Preferred reporting items for systematic reviews and meta-analysis 2009 flow diagram. Flowchart representing the search among different databases and included studies|
Click here to view
The studies showed that AI and machine learning models were able to predict the 5-year survival rate among oral cancer patients. The accuracy of the models such as decision tree classifier, logistic regression model, and boosted decision tree was 76%, 60%, and 88.7%, respectively. Hung et al. reported that the site of the tumor, its size, the age and time of diagnosis of tumor, as well as the year of diagnosis were the factors that contribute to the increased survival rate of oral cancer. Among these, the year of diagnosis plays a huge role in longer survival rate as, those who were diagnosed in the early stages tend to have longer survival up to 60.35 months than those diagnosed in the later stages who had a survival of 36 months.
| Discussion|| |
There are various subsets in AI, which includes machine learning, deep learning, and neural network. Data mining is an old terminology used before and is the one that bridges the gap between AI and statistics. It is the process of extraction of data from large data sets and this, it discovers patterns, thereby transforming the data into an understandable form. The subsets of AI are explained below.
Machine learning is a subset of AI that focuses on making the machines (computers) make any kind of decisions by feeding them with data. Deep learning handle large data sets and has decision-making capacities. Neural networks are a set of algorithms that compute signals via artificial neurons. The purpose of the neural network is to create neural networks that function like the human brain.
The primary goals of these studies were to fabricate a machine learning model for oral cancer patients, which would be useful for predicting the survival rate and also to show the key factors for survival prediction of oral cancer or mouth neoplasms. Oral cancer patients in these studies survived in an average of up to 60.35 months in recent times.
Karadaghy et al. created an survival prediction using machine learning with multiple variables and compared this with the prediction model, which included only the TNM clinical and pathologic stage in oral squamous cell carcinoma patients. The variables included in the machine learning were classified as the patient, facility, tumor, and treatment characteristics. Some of the variables were demographics, clinical, pathologic variables (size of the lymph node), age of the patient, and insurance status. Electronic databases of patients' health records were used to provide the patient's with an accurate prognosis. These databases were taken from the American population. However, the possibility of using these variables in the Indian scenario is not known. When comparing the two models, the accuracy for machine learning with multiple variables was 71%, whereas in TNM staging, the accuracy was only found to be around 65%. Thus for a more accurate prediction combination of multiple factors is required. And also to find these factors huge data are required.
The unique feature of machine learning is that it can apply Boolean logic. The machine learning algorithms used were logistic regression, support vector machine (SVM), naive Bayes, Neural network (NN), boosted decision tree, decision forest, decision jungle, random forest, k-nearest neighbors, single tree, and treeboost.
The data mining technique was used to solve association, classification, segmentation, diagnosis, and prediction problems. The term decision tree regression model determines the top variables used for survival prediction by using relative variable important scores and it was the ideal one. These methods are more suitable for clinicians, but these have poor performances. Random forest method consists of hundred tree-structured classifiers where each tree classifier casts votes, in which the outputs were shown. The SVMs come under the neural network (NN) subset. These models were in closely linked to neural networks. It is a two-layer, feed-forward network.
This will be useful for the clinicians to change their treatment plan accordingly and chances are being there that the patient's disease survival can be predicted. These models give more accurate outcomes of oral cancer patients better when compared with nomograms or other study models. Alabi et al. concluded that these models predicted the survival rate of oral cancer patients, which made the clinicians modify the treatment plan, thereby increasing the survival rate and also improving the quality of life.
Various models which have been used in the final six articles are explained as follows. Sharma and Om. concluded that the Tree boost model was better than the single tree and decision tree forest. In another study, they concluded that SVM is the most favorable model for predicting the survival rate of oral cancer patients. Karadaghy et al. concluded that decision forest, decision jungle, logistic regression, and neural network showed an accuracy of 71% than those which are calculated using the TNM staging system, which showed an accuracy of 65%. Alabi et al. concluded that more personalized and very reliable details of oral tongue cancer were provided by machine learning than nomogram. The models used in this study are logistic regression, SVM, Bayes point machine, boosted decision tree, decision forest, and decision jungle. Among these models, Boosted tree model performed better than all the other models. The nomogram produced an accuracy of about 60.4%, while the boosted tree model produced an accuracy of about 88.7%. Hung et al. used the models such as linear regression, decision tree, random forest, XG boost for the study. Among this, the XG boost machine learning algorithms showed the best performance in which the primary site of the tumor played as an important factor in oral cancer survival prediction followed by the year and age of diagnosis. Alkhadar et al. concluded that the model which performed the best was the decision tree classifier which displayed accuracy of 76%, followed by a logistic regression model, which showed an accuracy of 60%. The naive Bayes did not display any predictive value. Machine learning models were deployed readily in a clinical setting. These studies suggest that more than 50% of mouth neoplasms are lethal, but early detection is the key factor for increasing survivability.
The limitations of these models were that they lacked transparency in analysis and these studies were created using retrospective cohorts. Other limitations of these studies were that they did not incorporate the physiological and psychological factors which could have explained the quality of life of oral cancer survivors.
This systematic review concluded that the highest accuracy rate of oral cancer prediction was 88% and the survival period of oral cancer patients was around 5 years. Further research on AI on oral cancer survival prediction, including the physiological and psychological factors, has to be done to increase the survival period of oral cancer patients.
| Conclusion|| |
Human intelligence can never be completely replaced by AI or machine learning for clinical decision-making. Since the diagnosis and survival prediction for oral cancer can vary from person to person, the incorporation of clinical knowledge along with AI will be more accurate and it will also be useful for clinicians to change their treatment plans accordingly. This also gives a sense of hope to the patient. These studies are important for identification as well as for the prediction of oral cancer, which will contribute to future advancements in this field and reduce health-care problems all around the world.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Alkhadar H, Macluskey M, White S, Ellis I, Gardner A. Comparison of machine learning algorithms for the prediction of five-year survival in oral squamous cell carcinoma. J Oral Pathol Med 2021;50:378-84.
Kim DW, Lee S, Kwon S, Nam W, Cha IH, Kim HJ. Deep learning-based survival prediction of oral cancer patients. Sci Rep 2019;9:6994.
Hung M, Park J, Hon ES, Bounsanga J, Moazzami S, Ruiz-Negrón B, et al.
Artificial intelligence in dentistry: Harnessing big data to predict oral cancer survival. World J Clin Oncol 2020;11:918-34.
Kar A, Wreesmann VB, Shwetha V, Thakur S, Rao VU, Arakeri G, et al.
Improvement of oral cancer screening quality and reach: The promise of artificial intelligence. J Oral Pathol Med 2020;49:727-30.
Turner OC, Aeffner F, Bangari DS, High W, Knight B, Forest T, et al.
Society of toxicologic pathology digital pathology and image analysis special interest group article*: Opinion on the application of artificial intelligence and machine learning to digital toxicologic pathology. Toxicol Pathol 2020;48:277-94.
Moher D, Liberati A, Tetzlaff J, Altman DG; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. PLoS Med 2009;6:e1000097.
Sharma N, Om H. Data mining models for predicting oral cancer survivability. Netw Model Anal Health Inform Bioinform 2013;2:285-95.
Sharma N, Om H. Using MLP and SVM for predicting survival rate of oral cancer patients. Netw Model Anal Health Inform Bioinform 2014;3:58.
Alabi RO, Mäkitie AA, Pirinen M, Elmusrati M, Leivo I, Almangush A. Comparison of nomogram with machine learning techniques for prediction of overall survival in patients with tongue cancer. Int J Med Inform 2021;145:104313.
Sultan AS, Elgharib MA, Tavares T, Jessri M, Basile JR. The use of artificial intelligence, machine learning and deep learning in oncologic histopathology. J Oral Pathol Med 2020;49:849-56.
Karadaghy OA, Shew M, New J, Bur AM. Development and assessment of a machine learning model to help predict survival among patients with oral squamous cell carcinoma. JAMA Otolaryngol Head Neck Surg 2019;145:1115-20.
[Table 1], [Table 2], [Table 3]