|Ahead of print publication
The beginning of a new era: Artificial intelligence in oral pathology
C Nandini1, Shaik Basha2, Aarchi Agarawal2, R Parikh Neelampari3, Krishna P Miyapuram4, R Jadeja Nileshwariba3
1 Department of Oral Pathology and Microbiology, Gujarat University, Ahmedabad, Gujarat, India
2 Department of Computer Science Engineering, Indian Institute of Technology (IIT), Gandhinagar, Gujarat, India
3 Department of Oral and Maxillofacial Pathology, Karnavati School of Dentistry, Karnavati University, Gandhinagar, Gujarat, India
4 Centre for Cognitive and Brain Sciences, Indian Institute of Technology, Gandhinagar, Gujarat, India
|Date of Submission||16-Sep-2021|
|Date of Acceptance||08-Feb-2022|
|Date of Web Publication||23-Sep-2022|
4, New Pushpak Society, Behind Avsar Party Plot, Hansol, Ahmedabad - 382 475, Gujarat
Source of Support: None, Conflict of Interest: None
Intelligence is one of the vital qualities of the human brain, and it has been sometimes defined as the capacity to learn and understand new things, the ability to apply knowledge to manipulate one's environment or to think abstractly. When a machine is trained to learn and perform tasks in such a humanoid manner, it is referred to as the 'Artificial intelligence (AI)'. AI is likely to transform the way we live and work. The exponential growth of knowledge in the field of AI and its branches in the past decade has created new opportunities for its utilisation in the field of healthcare, including pathology. Successful application of these powerful tools in pathology and medicine requires cross-disciplinary literacy, including basic knowledge and understanding of concepts that have traditionally been unfamiliar to pathologists. This review provides definitions and basic knowledge of AI and its component branches such as machine learning, artificial neural networks and deep learning. This review also throws light on the possible applications of AI and the associated challenges in mainstream oral pathological research and diagnosis.
Keywords: AI in oral pathology, artificial intelligence, deep learning, machine learning
|How to cite this URL:|
Nandini C, Basha S, Agarawal A, Neelampari R P, Miyapuram KP, Nileshwariba R J. The beginning of a new era: Artificial intelligence in oral pathology. Adv Hum Biol [Epub ahead of print] [cited 2022 Sep 27]. Available from: https://www.aihbonline.com/preprintarticle.asp?id=356794
| Introduction|| |
Pathology diagnosis has been traditionally performed by pathologists observing the stained-glass slides under microscopes. In recent years, with the advent of digital pathology equipment such as microscopic cameras, high throughput scanning devices and whole slide imaging systems, computer-assisted diagnosis has become a reality.
Furthermore, the production of abundant digital pathological data has led to innovative research by application of artificial intelligence (AI) tools such as machine learning (ML) and deep learning (DL).
In this review, we present an overview of AI, a brief introduction to types of ML tools and their possible applications in oral pathology, along the associated challenges.
| What is Artificial Intelligence?|| |
AI (a term coined by John McCarthy in 1955) refers to the ability of machines to acquire and apply knowledge to perform a variety of cognitive tasks (e.g., sensing, processing oral language, reasoning, learning and making decisions), basically mimicking human behaviour. Key components included under the umbrella term AI have been summarised in [Figure 1]. The adoption of AI in the healthcare sector can be beneficial in early diagnosis, decision-making and treatment. Thus, it is expected to see an exponential increase in the next few years.
| Machine Learning|| |
ML (a term coined by Arthur Samuel, 1959) is a subset of AI that provides systems with the ability to automatically learn and improve from experience without being explicitly programmed. ML develops computer programmes that can access data and use it to learn for themselves.
| Types of Machine Learning|| |
Supervised ML algorithms need external assistance and training with labelled datasets. The collected data that is fed into the computer is called a dataset, composed of either discrete variables (requiring classification) or continuous variables (requiring regression). The components of the dataset are appropriately labelled and randomly divided into a training and test (validation) dataset. AI algorithms learn some kind of patterns from the labelled training dataset and apply them to the test dataset for prediction or classification. The workflow of supervised ML algorithms is schematically represented in [Figure 2]. The most commonly used supervised ML algorithms are diagrammatically represented in [Figure 3].
|Figure 2: Workflow of supervised and unsupervised machine learning algorithms.|
Click here to view
|Figure 3: Types of supervised and unsupervised Machine Learning Algorithms. (a) Decision tree; (b) NaÏve Bayes classifier; (c) Support Vector Machine; (d) Logistic Regression and Linear Regression; (e) k-means clustering; (f) Principal Component Analysis.|
Click here to view
In this method, an unlabelled training dataset is given to the learning algorithm, leaving it on its own to find structure in its input. When the new unlabelled testing dataset is introduced, it uses the previously learned features to recognise the class of the data. The workflow of unsupervised ML algorithms is schematically represented in [Figure 2]. It is mainly used for clustering and feature extraction. The most commonly used unsupervised ML algorithms are explained with a diagrammatic representation in [Figure 3].
Reinforcement learning takes actions such that the outcome is more positive. The learning algorithm has no knowledge of which actions to take until it's been given a situation.
| Neural networks|| |
The neural network (artificial neural network or [ANN]) is derived from the biological concept of neurons. A neuron has dendrites, nucleus, soma and axon. [Figure 4]a explains the structure of a neuron. The interconnection of neurons is called a neural network, where electrical impulses travel around the brain.
|Figure 4: Comparison between a biological neuron and artificial neuron. (a) Neuron: The dendrites receive electrical signals through synapse. Soma processes the electrical signal. Axon carries output of the process is to the next neuron through dendrite terminal. The nucleus is the heart of the neuron. (b) The input layer takes input, the hidden layer processes the input and the output layer sends the calculated output (c) Neural networks used for deep learning models have greater number of hidden layers than in a simple artificial neural network, and hence called ‘deep’ learning models.|
Click here to view
An ANN behaves the same way. It works on three layers. The input layer takes input (synapse and dendrites). The hidden layer processes the input (such as soma and nucleus) and the output layer sends the calculated output (such as axon and dendrite terminals). A comparison between a biological neuron and an artificial neuron is shown in [Figure 4]b. ANN models become increasingly complex with an increase in the number of hidden layers.
| Deep Learning|| |
DL, while sounding flashy, is really just a term to describe certain types of neural networks that often consume very raw input data. Since neural networks used for these models have a greater number of hidden layers than in simple ANN [Figure 4]c, they are called DL models. ANN and DL algorithms point towards a promising future of digital pathology as they are particularly useful in image recognition and classification.
| Applications of Machine Learning in Oral and Maxillofacial Pathology|| |
ML and DL algorithms are being increasingly utilised for automation of image analysis, computer-assisted diagnosis and content-based image retrieval., A brief review of the published literature related to applications of ML algorithms in oral pathology is summarised in [Table 1]. Applications of ML tools in oral and maxillofacial pathology are:
|Table 1: Brief review of the published literature related to applications of machine learning algorithms in oral pathology|
Click here to view
Machine learning in oral cytopathology
Since cytopathology involves a highly subjective evaluation of images of the cytosmears, it is suitable for the application of ML and ANNs for a more objective morphometric interpretation. The PAPNET™, an automated interactive system for conventional cervical smears, was used in the automation of oral cytopathology in 1998 and detected oral squamous cell cancer (OSCC) with an accuracy of 61%. More recently, Convolutional Neural Network (CNN), along with techniques such as mobile microscopes, telecytology and cytology-on-a-chip techniques, have proved to be accurate for image classification tasks involved in cytopathology.,,,
Computer-assisted diagnosis of oral cancer
Approximately two-thirds of the OSCC patients present with stage III or IV advanced disease, where surgical cancer resection with concurrent chemotherapy or radiation therapy remains the primary management. The primary goal of histopathology reporting of these resection specimens is to confirm tumour free and safe margins of surgical resection. This process involves complex grossing techniques, examination of a large number of histological slides and considerable diagnostic time.
To reduce the human labour and diagnostic time in reporting of these specimens, ML techniques have been used in diagnosing colorectal, breast, lung and in head-and-neck cancers with high accuracy. The digital workflow and steps involved in computer assisted diagnosis are shown in [Figure 5].
Lymph Node Assistant (a neural network for identifying cancer cells in lymph node biopsies developed by Google AI) and other algorithms have been used to identify lymph node micrometastasis.
Computer-assisted grading of oral epithelial dysplasia
Oral epithelial dysplasia (OED) has always been associated with a high interobserver and intraobserver disagreement. As a result, the need for computer-assisted grading and diagnosis of OED has been strongly advocated by oral pathologists to make the grading systems more reliable.
Several attempts have been made in the past to digitise the grading of OED using image analysis software. However, the use of the image analysis software in primary diagnosis without ML algorithms is not practical as it involves ample human effort and time.
Sami et al. in proposed to work on photomicrographs of oral epithelium. Several authors attempted classification of OED into mild, moderate and severe epithelial dysplasia and achieved an accuracy of around 92%.,,
Gupta et al. in aimed to classify OED into mild, moderate and severe epithelial dysplasia using a popular DL technique, namely CNN, with an accuracy of 89.3%.
Applicability of machine learning tools in scoring of immunohistochemical staining
IHC image analysis using DL techniques quantitatively estimates the disease-related protein expression and improves scoring reproducibility.
| Ethics Related to Using AI in Medicine and Pathology|| |
Usage of AI in medicine and pathology has given rise to a few questions. Who or what is the agent of responsibility? Neri et al. postulated that the human healthcare expert should be held responsible for what AI technology does. Thus, full automation is probably neither possible, and it seems, nor wise. Further, the clauses such as ownership and rights of the patients contributing to the dataset, ML engineers who build the algorithms and pathologists who perform the CAD remain unanswered.
| Conclusion|| |
Digital pathology and the research on the possible application of AI are growing at a tremendous rate and have the potential to transform pathology by increasing efficacy in diagnostics and the clinical outcome of patients. The field of pathology AI is still in its infancy and will mature as researchers, pathologists and patient groups work together to deliver precise and clinically applicable technologies. Despite the perceived threat of AI, it is plausible that AI tools that generate and analyse big image data will be a boon to pathologists by increasing their value, efficiency, accuracy and personal satisfaction.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Chang HY, Jung CK, Woo JI, Lee S, Cho J, Kim SW, et al.
Artificial intelligence in Pathology. J Pathol Transl Med 2019;53:1-12.
Serag A, Ion-Margineanu A, Qureshi H, McMillan R, Saint Martin MJ, Diamond J, et al.
Translational AI and deep learning in diagnostic pathology. Front Med (lausanne) 2019;6:185.
Mahind R, Patil A. A review paper on general concepts of artificial intelligence and machine learning. Int Adv Res J Sci Eng Technol 2017; 4:79-82.
Kohavi R, Provost F. Glossary of terms. Mach Learn 1998;30:271-4.
Dey A. Machine Learning Algorithms: A Review. International Journal of Computer Science and Information Technologies. 2016;7:1174-9.
Wang D, Khosla A, Gargeya R, Irshad H, Beck A.H, “Deep Learning for Identifying Metastatic Breast Cancer,” arXiv preprint arXiv: 2016;1606:05718.
Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z. TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems; 2015. Available from: http://tensorflow.org/
. [Last accessed on 2021 Sep 10].
Janowczyk A, Madabhushi A. Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases. J Pathol Inform 2016;7:29.
] [Full text]
Qi X, Wang D, Rodero I, Diaz-Montes J, Gensure RH, Xing F, et al.
Content-based histopathology image retrieval using CometCloud. BMC Bioinformatics 2014;15:287.
Levine TS, Njemenze V, Cowpe JG, Coleman DV. The use of the PAPNET automated cytological screening system for the diagnosis of oral squamous carcinoma. Cytopathology 1998;9:398-405.
Forslid G, Wieslander H, Bengtsson E, Wählby C, Hirsch JM, Stark CR, et al
. Deep Convolutional Neural Networks For Detecting Cellular Changes Due To Malignancy. IEEE International Conference on Computer Vision Workshops (ICCVW), Venice; 2017. p. 82-9.
Skandarajah A, Sunny SP, Gurpur P, Reber CD, D'Ambrosio MV, Raghavan N, et al.
Mobile microscopy as a screening tool for oral cancer in India: A pilot study. PLoS One 2017;12:e0188440.
McRae MP, Modak SS, Simmons GW, Trochesset DA, Kerr AR, Thornhill MH, et al
. Point-of-care oral cytology tool for the screening and assessment of potentially malignant oral lesions. Cancer Cytopathol 2020;128:207-20.
Helicek M, Maysam S, James VL, Amy YC, Larry LM, Baran DS, et al.
Head and neck cancer detectionin digitized whole-slide histology using convolutional neural networks. Sci Rep 2019;9:14043.
Xu J, Luo X, Wang G, Gilmore H, Madabhushi A. A deep convolutional neural network for segmenting and classifying epithelial and stromal regions in histopathological images. Neurocomputing 2016;191:214-23.
Hou L, Samaras D, Kurc TM, Gao Y, Davis JE, Saltz JH. Patch-based Convolutional Neural Network for Whole Slide Tissue Image Classification. Proceedings of the IEEE conference on computer vision and pattern Recognition (CVPR). 2016;2016:2424-33. doi: 10.1109/CVPR.2016.266. PMID: 27795661; PMCID: PMC5085270.
Steiner DF, MacDonald R, Liu Y, Truszkowski P, Hipp JD, Gammage C, et al.
Impact of deep learning assistance on the histopathologic review of lymph nodes for Metastatic breast cancer. Am J Surg Pathol 2018;42:1636-46.
Sami MM, Saito M, Muramatsu S, Kikuchi H, Saku T. A computer-aided distinction method of borderline grades of oral cancer. IEICE Trans Fundam Electron Commun Comput Sci 2010;93:1544-52.
Aurchana P, Dhanalakshmi P, Chidambaram C. SVM based classification of epithelial dysplasia using surf and sift features. Int J Pure Appl Math 2017;17:1163-75.
Krishnan MM, Chakraborty C, Paul RR, Ray AK. Hybrid segmentation, characterization and classification of basal cell nuclei from histopathological images of normal oral mucosa and oral submucous fibrosis. Expert Syst Appl 2012;39:1062-77.
Gupta RK, Kaur M, Manhas J. Tissue level based deep learning framework for early detection of dysplasia in oral squamous epithelium. J Multimed Inf Syst 2019;6:81-6.
Tizhoosh HR, Pantanowitz L. Artificial intelligence and digital pathology: Challenges and opportunities. J Pathol Inform 2018;9:38.
] [Full text]
Neri E, Coppola F, Miele V, Bibbolino C, Grassi R. Artificial intelligence: Who is responsible for the diagnosis? Radiol Med 2020;125:517-21.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]