The scientific study of disease and its mechanisms, known as pathology, has intrigued scientists since 1700 BC.^1,2 In the past, pathology involved the manual inspection of biopsies (tissue removed from a living organism) a crucial but time-consuming and subjective task that could result in variations in diagnosis and affect patient treatment.² The emergence of Artificial Intelligence (AI) is set to transform this area by improving the precision and speed of disease diagnosis.¹

Understanding AI in pathology

AI a discipline focused on imitating human intelligence using technology can be traced back to the 1930s with the development of Alan Turin’s Turing machine.¹ Advancement in AI was restricted due to the lack of large-scale data, limited computational power and insufficient research funding.² However, with a growing interest in the use of AI in science and medicine, these limitations have been progressively overcome in the last decade, placing AI at the forefront of technological innovation.³ Today, AI encompasses a range of sophisticated techniques, including machine learning, deep learning and natural language processing (NLP).³ 

Machine learning involves algorithms that learn from the data and apply this knowledge to new, unseen data to make decisions.³ Deep learning, a branch of machine learning, uses layered structures called neural networks to simulate human decision-making processes.³ Among AI’s many applications, NLP has become particularly well-known to the general public through AI tools like ChatGPT, which can understand and generate human language responses.⁴ These AI methods require a vast amount of data to improve accuracy, enabling machines to learn from data, make predictions, and perform tasks once reliant on human intelligence.³

AI in pathology is increasingly being applied for a variety of uses in image analysis, demonstrating impressive success in streamlining diagnoses.¹ The most popular method is deep learning, which uses prior knowledge developed from training data.³ The data includes vast amounts of pathological information obtained from quality-controlled patient samples.¹ The accuracy of AI models depends heavily on the volume and representativeness of the training data, ensuring it reflects various disease subtypes to avoid bias.⁵ For example, cancers are categorised into different grades and stages based on different features of the tissue on the slides.⁶ Thus, both the diversity and quantity of training data are crucial for accurate, unbiased disease diagnosis.⁶ Understanding the specificities of AI application in pathology is crucial as it highlights the importance of data quality and diversity, ensuring that AI models can provide accurate and reliable diagnostics across diverse patient populations.

Applications of AI in pathology

Histopathological image analysis

One prominent application of AI in pathology is in the analysis of histopathological images, particularly in cancer diagnosis. Histopathology involves studying changes in tissue samples caused by disease.7 The assessment of histopathological tumour samples involves several specific steps, including the preparation of image slides and subsequent histological characterisation by trained pathologists.8 This characterisation includes examining different cell types in the sample, quantifying these cells and analysing other features that help determine the cancer’s grade and stage.^7,8 For instance, AI systems can accurately identify breast cancer cells in whole-slide images, significantly aiding pathologists in diagnosing and assessing the extent of the disease​.⁷

Prognosis and treatment response prediction

AI in pathology is also utilised to predict outcomes and treatment responses in cancer.⁵ Histopathological samples contain features that can serve as prognostic biomarkers, such as the number of specific immune cell types in the sample or the state of the genetic material within the cells.⁹ AI’s ability to predict treatment responses is of particular interest, as it could assist healthcare professionals in making better treatment recommendations.⁸ Some studies have identified cellular markers in colorectal or endometrial cancer that make these types susceptible to specific immunotherapeutic treatments using certain immune checkpoint inhibitors.⁸ Although these AI models require a long-term follow-up study and are currently limited in use, they show great promise for the future.⁸

Rare disease detection  

Another crucial application of AI in pathology is detecting rare diseases.¹⁰ AI algorithms can process vast amounts of data and recognise subtle anomalies that might be missed by humans.¹¹ Moreover, AI's integration with electronic health records (EHRs) allows it to correlate rare clinical presentations with underlying pathological features, thus improving the diagnostic workflow and enabling timely intervention.¹¹ As AI technologies advance, their applications in detecting and diagnosing rare diseases are expected to become even more robust, providing critical support to pathologists and improving patient outcomes through earlier and more accurate diagnoses.¹⁰

Workflow optimisation

Workflow optimisation is another area where AI in pathology offers significant benefits​.^12,13 For example, AI algorithms trained to pre-screen slides can identify normal versus abnormal tissue samples, ensuring that only those requiring further examination by a pathologist are flagged.¹³ This approach simplifies and accelerates the diagnostic process, allowing expert pathologists to focus on complex cases that demand their expertise​​.¹² By streamlining processes and enhancing diagnostic accuracy, AI helps address the shortage of pathologists and ensure timely, precise patient care​.

Reducing human error

AI also significantly reduces human error in pathology​.¹³ Automated systems minimise the variability associated with manual interpretation, resulting in more consistent and reliable diagnoses​.(14) AI’s rapid image processing capabilities allow for the quick analysis of large volumes of data, facilitating timely treatment interventions crucial for conditions like cancer, where early diagnosis can significantly impact patient outcomes.^13,14

Cost-effectiveness

Furthermore, AI can be cost-effective by optimising resource use and reducing the need for repeat tests.¹⁵ In the UK healthcare system, this can translate into significant savings and more efficient use of NHS resources.^15,16 Integrating AI into pathology workflows can alleviate some of the burdens on the NHS by enhancing diagnostic efficiency and accuracy, ultimately contributing to better patient care and resource management​.¹⁶ As these technologies continue to evolve, their integration into pathology workflows is expected to become more widespread, further revolutionising the field and improving patient outcomes.^15,16

Challenges and limitations

Despite the promising potential of AI in pathology, several challenges and limitations restrict its widespread adoption.¹⁵ One major issue is the need for large, annotated datasets to train AI algorithms effectively.^13,15 These datasets must represent diverse populations and disease subtypes to avoid biases that could lead to inaccurate diagnoses.¹⁵ Additionally, integrating AI systems into existing pathology workflows can be complex and costly, requiring significant investment in digital infrastructure and training for pathologists and technicians.¹³

Another notable limitation is the interpretability of AI models, in deep learning systems, where the “black box” problem arises.This term refers to the difficulty in understanding how AI systems combine variables to make predictions, as it often involves thousands of connections within the data network​.¹² This lack of transparency can reduce trust in AI-generated results and complicate the validation process required for clinical use​.¹⁶

Regulatory and ethical considerations also pose significant challenges.¹³ Implementing AI in clinical settings must comply with stringent regulatory standards to ensure patient safety and data privacy.¹⁷ In the UK, for example, the National Health Service (NHS) must navigate these regulatory landscapes while ensuring that AI tools meet the high standards required for medical applications​.^17,18

Furthermore, the potential for AI to reduce human error and streamline workflows must be balanced against the risk of over-reliance on automated systems.¹⁹ Pathologists may become dependent on AI, potentially reducing their diagnostic skills over time.¹⁹ Ensuring a balanced integration where AI acts as a supportive tool rather than a replacement for human expertise is crucial​.

Overall, while AI holds significant promise for enhancing pathology practices, addressing these challenges through comprehensive datasets, transparent algorithms, regulatory compliance, and balanced integration is essential for its successful implementation.

Future prospects and conclusion

The future of AI in pathology holds great promise, with potential advancements that could further revolutionise the field. As AI technology continues to evolve, its applications in pathology are expected to expand, offering even more sophisticated tools for disease detection, diagnosis, monitoring, and treatment planning.⁵ Future developments will likely enhance the accuracy and robustness of AI models through the use of more diverse and comprehensive datasets. Improving the transparency and interpretability of AI algorithms will also be crucial for building trust among pathologists and clinicians​.

One of the most exciting prospects is AI's potential to facilitate personalised medicine by integrating various types of patient data, including genetic, molecular, and histopathological information, to provide tailored treatment recommendations.²⁰ This capability could revolutionise disease diagnosis and treatment, leading to improved patient outcomes and more efficient healthcare delivery​.

To ensure successful implementation of AI in pathology and disease diagnosis, continued research and development is essential. This will require collaborative efforts between researchers, healthcare providers and regulatory bodies to address the existing limitations and ensure the safe application of AI tools.

In conclusion, AI has the potential to transform pathology by improving diagnostic accuracy, efficiency, and patient care. The promise of AI in pathology underscores the importance of continued innovation and commitment to harnessing technology for the betterment of healthcare systems worldwide.