Artificial intelligence (AI) involves the simulation by computer systems of the decision processes of the human brain, with the aim of mimicking certain actions such as problem-solving and learning. AI systems have been shown to be able to perform tasks that generally require human intelligence and in recent years, AI-based computational pathology has started to yield promising results. AI also has the ability to handle an enormous quantity of patient-related data and can improve diagnosis, prediction, classification, and the prognostication of diseases.

AI can currently be divided into two categories: weak AI and strong AI. Weak AI involves classification of the data based on a statistical model that has already been trained to perform a single task. Strong AI involves creation of a system that is capable of determining multiple potential solutions from a single set of observations, in the same way as a human brain.

Role of AI in the diagnostic processes

Artificial intelligence (AI) has already had a major impact on the field of medicine, especially in diagnosis. Nowadays due to the increase in treatment options, much more accurate diagnostics are needed to meet the requirements of precision medicine. AI systems can even detect parameters that cannot be determined by human visualization, so AI can potentially bring both higher accuracy and reproducibility to the diagnostic process.

AI has proved particularly useful with histopathological tissue sections, which require specialized analysis by trained pathologists. The pathologist’s role is to diagnose and grade diseases such as inflammatory diseases or cancer based on an observation of tissue features. We are currently witnessing a shortage of trained pathologists and AI can be used to reduce some of the more time consuming and tedious tasks such as counting cells and screening for specific morphological characteristics.

AI in tissue analysis is often termed computational pathology. One of the most important reasons for the implementation of computational pathology in histopathology is that the algorithms can quickly yield powerful feature representations that are superior to traditional image analysis methods.

Advancement of computational pathology

over the last ten years, there has been considerable progress on multiple fronts with improvements in microscopic scanning devices, in the acquisition of whole-slide images (WSI), a decrease in the cost of hardware, and further advances in AI.

In 2016, computational pathology solutions were deployed in the CAMELYON challenge to look at the detection of breast cancer metastases in sentinel lymph nodes. The challenge was to find tumor regions in the lymph node and predict the presence of the tumor at the WSI level. CAMELYON attracted a large number of entrants as well machine learning powerhouses such as Google.

Role of computational pathology in clinical practice

AI algorithms not only help to automate existing diagnostic tasks but also provide pathologists with additional information, such as highlighting regions of prostate cancer with different colors to show different Gleason grades. Computational Pathology has also been deployed in colorectal cancer to distinguish between five types of colorectal polyps that involve sessile serrated, hyperplastic, traditional serrated, villous, and tubular polyps.

The outbreak of COVID-19 has led many AI companies to work on products that could strengthen control mechanisms in this type of public health emergency. For example, Schaar et al. have noted that machine learning has the capability to enhance the effectiveness as well as efficiency of randomized clinical trials for COVID-19.

Computational pathology also utilizes a combination of segmentation, detection, and classification that can assist in the quantification of classical biomarkers used in clinical practices. Assessment of tumor-infiltrating lymphocytes can be done by segmenting stromal regions of a slide and detecting intrastromal lymphocytes either by immunohistochemistry (IHC) or by hematoxylin and eosin (H&E) staining.

Challenges

Although there have been huge strides forward in computational pathology, specifically in the last 5 years, many challenges still exist. For one, although collecting a large number of WSIs can be carried out by medical centers and pathological laboratories, collecting enough manual annotations for the large scale validation of these algorithms remains a significant hurdle.

Secondly, although the size of the datasets for the development of computational pathology algorithms have grown substantially, they do not represent the type of data encountered in clinical practice. Also, most current solutions are point solutions (which can be defined as weak AI). In this case, the algorithm only recognizes the morphology it was trained for and not any other patterns.

Thirdly, the use of patient data and the deployment of machines which are assisting diagnosis can lead to ethical concerns. The use of human data for the development of products as well as health care research involves legal and ethical challenges that need to be properly addressed.

Perspectives

Application of AI in histopathological images has resulted in computational pathology solutions whose performance is comparable to a trained pathologists for specific defined diagnostic tasks. Currently, the use of AI in the discovery of prognostic features, the assessment of the relationship between the morphological phenotypes of disease and genotype, or prediction of therapy success is still being explored.

Although there are certain challenges, we can expect computational pathology to play a dominant role in the future of histopathology, helping pathologists to meet the needs of increased numbers of patients as well as making diagnostic decision making more accurate and efficient.

References

1. Cui M, Zhang DY. Artificial intelligence and computational pathology. Laboratory Investigation (2021) 101:412–422. https://doi.org/10.1038/s41374-020-00514-0.
2. Laak J, Litjens G, Ciompi F. Deep learning in histopathology: the path to the clinic. Nature Medicine VOL 27 May 2021 775–784. https://doi.org/10.1038/s41591-021-01343-4 .