|Year : 2022 | Volume
| Issue : 2 | Page : 117-120
Artificial intelligence and machine learning in head and neck oncology
Department of Head and Neck Surgery and Oncology, Amrita Institute of Medical Sciences, Kochi, Kerala, India
|Date of Submission||24-Nov-2022|
|Date of Acceptance||24-Nov-2022|
|Date of Web Publication||15-Dec-2022|
Department of Head and Neck Surgery and Oncology, Amrita Institute of Medical Sciences, Kochi, Kerala
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
Thankappan K. Artificial intelligence and machine learning in head and neck oncology. J Head Neck Physicians Surg 2022;10:117-20
| Introduction|| |
Machine learning (ML) is an academic discipline that allows the computer to perform complex tasks that may not be humanly possible. It has elements of mathematics, statistics, and computer science. Recently, it has been employed in different fields, both academia and industry, to solve complex problems and develop products. It has tremendous potential in the field of medical science. This editorial aims to give an overview and introduce the discipline and its applications in head-and-neck oncology.
| Artificial Intelligence, Machine Learning, and Deep Learning|| |
ML and artificial intelligence (AI) are not the same. However, they are closely linked. ML is regarded as a subset of AI. The utilization of a computer system to imitate human cognitive processes, such as learning and problem-solving, is known as AI. It simulates human decision-making and learns from new knowledge using logic and mathematics. The science of creating computers and robots with intelligence that mimics and exceeds that of humans is known as AI. Programs with AI capabilities can contextualize and analyze data to deliver information or automatically initiate operations without human intervention. ML learns from data and applies mathematical models to it. In ML, algorithms are designed and used to draw conclusions from previous instances. If a behavior has occurred in the past, you can anticipate whether it will do so in the future. Historical data are used to provide accurate outcomes. To forecast sensible outputs, ML algorithms use computer science and statistics. This branch of AI uses algorithms to discover patterns automatically and acquire insights from data. Algorithms are collections of mathematical operations that describe the relationship between the variables.
Deep learning (DL) is a subset of ML using “neural” networks in layers. These neural networks attempt to simulate the human brain. DL drives many AI applications and services that improve automation, doing analytical tasks without human intervention. DL differs from classical ML in the type of data it uses and the learning methods. ML algorithms use structured labeled data to make predictions. The input data for the model are defined and organized into tables. DL algorithms can take unstructured data such as text and images. It automates feature extraction without the need for human experts.
| Supervised and Unsupervised Machine Learning|| |
Supervised ML refers to techniques in which a model is trained on a range of input variables (or features) associated with a known outcome. In the head and neck, training a model to relate a patient's characteristics (e.g., smoking status) or clinical factors (T stage, N stage) to predict a disease recurrence outcome (recurrence occurred or not) come under this category. The trained algorithm can make outcome predictions when applied to new data. It is called a classification model when the predictions are discrete or categorical (e.g., positive or negative, malignant or not). When the predictions are continuous (e.g., a score ranging from 0 to 100), it is referred to as a regression model.
Unsupervised ML does not involve a predefined outcome. Algorithms identify patterns without any prior inputs. Unsupervised learning does not require labeled data sets; instead, it detects patterns in the data. Unsupervised methods are so exploratory to find undefined patterns. They use dimension reduction techniques like clustering to identify clusters, leading to outcomes automatically. For example, genomic and precision medicine studies investigate outcomes without predefined outcome data. [Figure 1] shows the relationship between AI, ML, and DL.
|Figure 1: Relationship between artificial intelligence, machine learning, and deep learning|
Click here to view
| Steps of Machine Learning|| |
- Cleaning and organizing the data (to identify and correct problems such as outliers, missing data, and restructuring)
- Exploratory data analysis (statistical analysis, univariate/bivariate analysis, visualizations to understand the characteristics of the data, including the distributions and correlations between the variables.)
- Scaling (to make the continuous variables on the same scale)
- Training (usually done using 80% of the data set)/cross-validation (for hyperparameter tuning)
- Testing (usually done on 20% of the data set)
- Model evaluation and comparisons
- External validation.
| Model Evaluation l Comparisons|| |
When multiple models are tested, the models are compared using model evaluation metrics. In classification models, the metrics are mainly based on the confusion matrix. The important ones are accuracy, precision, recall, and F1 score. The receiver operating characteristic area under the curve (AUC) can also be used. Regression models use the mean absolute error, mean squared error (MSE), root MSE, and R-squared and adjusted R-squared. The definition of these terms and how they are used for the evaluation of models may be out of the purview of this article and hence not attempted. The best among the models is selected.,
| Computer Programming Languages and Machine Learning Algorithms|| |
Although many are available, Python and R are the most commonly used, and both are open-source tools. The common libraries used in Python are NumPy, Pandas, and Matplotlib. R statistical programming language also has many packages dedicated to ML. Commonly used algorithms are linear regression, logistic regression, decision tree, support vector machine, Naive Bayes, kNN, K-means, random forest, dimensionality reduction algorithms, and gradient boosting algorithms (Gradient boost (GBM), XGBoost, LightGBM, and CatBoost).
| Applications in Head-and-Neck Oncology|| |
Massive volumes of health data and the need to draw inferences and derive insights from these data have sparked considerable interest in ML. Among the most frequently mentioned medical applications of ML include improved cancer diagnosis and prognosis prediction by combining clinical and genomic data and adaptive clinical trial designs. Computer vision algorithms that enable the rapid identification of radiographic anomalies, delineation of surgical anatomy, and classification of malignant tissues in pathologic specimens (e.g., intraoperative frozen sections and fine-needle aspirate samples) at a speed comparable to or exceeding that of human operators are another promising application.
In general, the applications of ML in medicine are in (a) diagnosis, (b) prognostication, and (c) treatment.
| Diagnostic Applications|| |
Early detection of oral cancer in low-resource settings necessitates a point-of-care screening tool that empowers frontline health workers (FHW). A recent study by Birur et al. validated the accuracy of convolutional neural network (CNN) enabled mobile health device deployed with FHWs for delineation of suspicious oral lesions (malignant/potentially malignant disorders). Applications of ML include classifying malignant tissue based on radiologic (radiomics) and histopathologic features (pathomics). “Radiomics” is the broad term referring to “high-throughput extraction of quantitative features that result in the conversion of images into mineable data and the subsequent analysis of these data for decision support.” Radiomic approaches perceive medical imaging as a rich source for mining quantitative features as opposed to pictures intended solely for visual and subjective interpretation. ML is commonly used for predictive modeling in radiomics. An automated classification neural network to detect differences between normal and malignant head-and-neck tissues using hyperspectral imaging (HSI) is reported. HSI is a noninvasive imaging modality that captures reflected transmitted and fluorescent light from tissues. Previously validated for other solid organ/cutaneous malignancies, it is being validated in the head and neck.
ML has applications in assessing HPV status. With a comprehensive feature selection process, a general linear model was built to predict HPV status using the radiomic features. The selected features indicated that HPV-positive tumors are smaller and more spherical than HPV-negative tumors. ML is also tried to identify nodal metastasis and extracapsular spread using dual-energy computed tomography, distinguishing head-and-neck squamous cell carcinoma (SCC) from lymphoma, inflammatory, and normal nodes. ML is also studied in conjunction with magnetic resonance imaging (MRI)/ultrasound/positron emission tomography. In histopathological diagnosis, ML aids in digital pathology for the diagnosis of normal versus malignant tissue, classifying SCC across four Broder's grading with an accuracy of 97.5%; however, its clinical use for grading still needs to be proven. ML has been studied for the quantification of biomarkers.
| Prognostic Applications|| |
ML has been studied for prognostication by gathering clinical and pathological data and forming a predictive model that analyzes data to predict outcomes. Survival prediction models have been developed for oral, larynx, hypopharynx, and nasopharynx cancers. Prediction of locoregional recurrences in oral tongue SCC is studied using the pT stage, pN stage, and adverse pathological features with a specificity of 98% and an accuracy of 88%. Application of DL over traditional learning to predict distant metastases improved AUC 0.88–0.92. ML is used for the prediction of progression-free survival in advanced or unresectable head and neck cancer (HNC). Two-dimensional computed tomography texture analysis showed an association with overall survival in locally advanced head-and-neck squamous cell carcinoma treated with induction chemotherapy. It also has applications in nasopharyngeal cancers along with MRI for pretreatment risk assessment of distant metastases.
| Treatment Applications|| |
The promising application of ML in treatment is for automated radiotherapy planning. ML plays a role in delineating organs at risk, and the dose received, traditionally done by going through each section, making it time-consuming and cumbersome. Babier et al. designed a principal component analysis algorithm (i.e., a dimension reduction approach) to predict dose-volume histograms in oropharyngeal cancer, which were input into an optimization pipeline to generate radiotherapy plans automatically. Three-dimensional CNN has been trained for primary gross tumor volume contouring and adaptive radiotherapy contouring.
| Future|| |
AI and ML have the potential to become an integral component of oncology management decisions. They can affect every aspect of management, including screening, diagnosis, risk stratification, treatment planning, follow-up, and policy decisions. Through the identification of complex patterns, AI provides excellent opportunities for automating tasks. In this regard, research is essential to facilitate the interdisciplinary incorporation of these techniques, and future advancements in this field may pave the way for additional research. Before decisions are made solely based on AI algorithms, they must be thoroughly verified.
This material has never been published and is not currently under evaluation in any other peer-reviewed publication.
Not applicable as this is an editorial article with no patients involved.
Not applicable as this is an editorial article with no patients involved.
| References|| |
Sidey-Gibbons JA, Sidey-Gibbons CJ. Machine learning in medicine: A practical introduction. BMC Med Res Methodol 2019;19:64.
Sabharwal R, Miah SJ, Fosso Wamba S. Extending artificial intelligence research in the clinical domain: A theoretical perspective. Ann Oper Res 2022:1-32.
Thomas LB, Mastorides SM, Viswanadhan NA, Jakey CE, Borkowski AA. Artificial intelligence: Review of current and future applications in medicine. Fed Pract 2021;38:527-38.
Egger J, Gsaxner C, Pepe A, Pomykala KL, Jonske F, Kurz M, et al.
Medical deep learning-A systematic meta-review. Comput Methods Programs Biomed 2022;221:106874.
Birur N P, Song B, Sunny SP, Keerthi G, Mendonca P, Mukhia N, et al.
Field validation of deep learning based point-of-care device for early detection of oral malignant and potentially malignant disorders. Sci Rep 2022;12:14283.
Leite AF, Vasconcelos KF, Willems H, Jacobs R. Radiomics and machine learning in oral healthcare. Proteomics Clin Appl 2020;14:e1900040.
Gupta R, Kurc T, Sharma, A, Almeida JS, Saltz J. The emergence of pathomics. Curr Pathobiol Rep 2019;7:73-84.
Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images are more than pictures, they are data. Radiology 2016;278:563-77.
Halicek M, Lu G, Little JV, Wang X, Patel M, Griffith CC, et al.
Deep convolutional neural networks for classifying head and neck cancer using hyperspectral imaging. J Biomed Opt 2017;22:60503.
Yu K, Zhang Y, Yu Y, Huang C, Liu R, Li T, et al.
Radiomic analysis in prediction of human papilloma virus status. Clin Transl Radiat Oncol 2017;7:49-54.
Forghani R. An update on advanced dual-energy CT for head and neck cancer imaging. Expert Rev Anticancer Ther 2019;19:633-44.
Das N, Hussain E, Mahanta LB. Automated classification of cells into multiple classes in epithelial tissue of oral squamous cell carcinoma using transfer learning and convolutional neural network. Neural Netw 2020;128:47-60.
Kather JN, Schulte J, Grabsch HI. Deep learning detects virus presence in cancer histology. BioRxiv 2019:690206
Alabi RO, Elmusrati M, Sawazaki-Calone I, Kowalski LP, Haglund C, Coletta RD, et al.
Machine learning application for prediction of locoregional recurrences in early oral tongue cancer: A Web-based prognostic tool. Virchows Arch 2019;475:489-97.
Diamant A, Chatterjee A, Vallières M, Shenouda G, Seuntjens J. Deep learning in head & neck cancer outcome prediction. Sci Rep 2019;9:2764.
Zhao Z, Li Y, Wu Y, Chen R. Deep learning-based model for predicting progression in patients with head and neck squamous cell carcinoma. Cancer Biomark 2020;27:19-28.
Zhong J, Frood R, Brown P, Nelstrop H, Prestwich R, McDermott G, et al.
Machine learning-based FDG PET-CT radiomics for outcome prediction in larynx and hypopharynx squamous cell carcinoma. Clin Radiol 2021;76:78.e9-78, e17.
Babier A, Boutilier JJ, Sharpe MB, McNiven AL, Chan TC. Inverse optimization of objective function weights for treatment planning using clinical dose-volume histograms. Phys Med Biol 2018;63:105004.
Rattan R, Kataria T, Banerjee S, Goyal S, Gupta D, Pandita A, et al.
Artificial intelligence in oncology, its scope and future prospects with specific reference to radiation oncology. BJR Open 2019;1:20180031.