Introduction
Artificial intelligence (AI) denotes the intelligence offered by computers and robots. The AI process is teaching a machine to solve a problem without specifying each step to the solution. Machine learning (ML; gives the computers learning ability) and deep learning (DL; complex architectures similar to the brain neurons) are the two approaches to implementing AI.
Weak AI is designed to carry out a particular job (video games). On the other hand, strong artificial intelligence systems complete human-like tasks. These are more intricate systems (self-driving cars or in hospital operating rooms). Pathology is the scientific branch that plays a crucial role in disease diagnosis, drug development process, and preclinical research. While AI benefits many areas of clinical health sciences, AI in pathology includes education, quality assurance, and clinical diagnosis. Furthermore, AI has a meaningful and measurable impact on clinical and research components.
What Are the Uses of Artificial Intelligence in Pathology?
Since the initial stage of AI, substantial efforts have been made for medical research. The various applications of AI in pathology include:
Image Analysis (Digital Pathology): Microscopic morphology (study of form and structure) is the gold standard in pathology. However, the main limitation of morphologic diagnosis is error variation among pathologists. AI has been introduced in pathology for the morphological analysis of tissues and cells to get a consistent and accurate diagnosis.
Digital pathology refers to the digital image creation and analysis from scanned pathology slides. It is an essential field of AI research for early diagnosis. Digital pathology equipment comprises microscopic cameras, whole slide imaging scanners, and morphology-based automated pathologic diagnosis. AI focuses on morphology-based pathology: diagnosis and prognosis based on pathology image assessment. Typical digital image analysis in diagnostic pathology involves segmentation, detection, classification, quantification, and grading.
AI-assisted image analysis is also paramount for prognosis and prediction of treatment response. However, additional immunohistochemical (IHC; a staining method) or next-generation sequencing (a type of molecular biology) is used to determine prognostic and predictive biomarkers. AI algorithms can identify structural characteristics which the human eye cannot. Various studies describe DL for colorectal, prostate cancer, and other conditions. DL can also detect genetic mutation in gastrointestinal tumors and identify the molecular subtype of bladder cancer.
-
Genetic Data Analysis: Genomics (the study of the genetic information of a person) and molecular biology are the branches of pathology. AI plays an increasingly important role in molecular pathological data analysis. AI can also detect changes in morphology in tumor tissues. AI-based methods can make predictions directly from deoxyribonucleic acid (DNA) or protein sequence data. AI can also classify DNA profiles using DL to distinguish a metastasis (spread of cancer cells) of a head and neck carcinoma from lung cancer. DL-based approaches have improved the ability to predict the influence of genetic variation on molecules. Also, AI methods play a vital role in molecular tumor classifications. However, immediate clinical use has not been established. Still, evidence of the future significance of complex molecular and morphological profiles is emerging. Genetic syndromes can also be confirmed with DNA testing with AI.
-
Early Cancer Diagnosis: Diagnosing early cancer increases the chances of performing effective treatment in many tumors. AI algorithms can assist doctors by analyzing health records, medical images, biopsy samples, and blood tests to improve early diagnosis. Studies reveal that AI can predict pancreatic cancer risk using clinical parameters such as age, smoking, alcohol, and ethnicity. AI helps automate the detection and classification of precancerous lesions and early cancers. Various early detection approaches utilize high-dimension data for enhancement with AI.
Another application of AI to cancer research is earlier recurrence detection following treatment. It is because accurate detection can facilitate personalized therapy. Hence, the high-risk cases can be offered intensive primary treatment (for example, radiotherapy dose escalation). On the other hand, low-risk patients can receive less intensive treatment to reduce side effects.
Post-treatment follow-up is a recommended aspect of cancer care. It supports patients with treatment-related side effects, reassurance, and management of comorbidities. It can further offer earlier treatment for recurrence and improve the early diagnosis of secondary cancers. Studies show that ML with routine clinical data (patient, tumor, and treatment characteristics) can predict bladder cancer recurrence.
What Are the Challenges Faced by Artificial Intelligence in Pathology?
AI faces several challenges, including ethical considerations, algorithmic legitimacy, data bias, governance, and security. Developing ethical principles is the subject of healthcare AI. The World Health Organization (WHO) has probed healthcare AI contributors to ensure that newer technologies give ethics and human rights primary importance. The common concerns are the nature of AI decisions, the impact on patient experience, and shared decision-making.
A crucial concern is model bias (bias on demographic characteristics such as gender and ethnicity). In a study, an AI tool diagnosed skin cancer with about one lakh clinical images with less than five percent of darker skin images. As a result, it drew criticism about its reproducibility and validity. Hence, understanding and addressing structural racism can improve model accuracy and external validity. One must hope that measures to describe ethnic distributions will become adopted.
Further, data security is an ongoing concern due to recent leaks and the threat of inference attacks. Approaches are growing to improve data security and reduce the risks associated with transferring data. Also, data storage can be time-consuming and costly. Hence, many agencies require clear plans for data management. The requirement for large sets of labeled data for training presents a significant challenge to researchers.
Conclusion
Artificial intelligence in pathology is expanding to disease severity assessment and prognosis prediction. However, it still focuses on cancer detection and tumor grading. Artificial intelligence-aided pathological diagnosis is a complex process of evaluation and judgment of various data dealing with diseases. Although artificial intelligence and machine learning can revolutionize pathology, there are several challenges to their implementation. There is a strong relation between artificial intelligence algorithms and the amount of data. Hence, a new tumor grading system can be created for the patient’s prognosis combined with the morphology, treatment modality, and tumor markers. It will overcome the poor reproducibility of current grading and staging systems among pathologists. As a result, one can expect better clinical outcomes for patients.
