Expanding role and scope of artificial intelligence in the field of gastrointestinal pathology

Abstract

Digital pathology (DP) and its subsidiaries including artificial intelligence (AI) are rapidly making inroads into the area of diagnostic anatomic pathology (AP) including gastrointestinal (GI) pathology. It is poised to revolutionize the field of diagnostic AP. Historically, AP has been slow to adopt digital technology, but this is changing rapidly, with many centers worldwide transitioning to DP. Coupled with advanced techniques of AI such as deep learning and machine learning, DP is likely to transform histopathology from a subjective field to an objective, efficient, and transparent discipline. AI is increasingly integrated into GI pathology, offering numerous advancements and improvements in overall diagnostic accuracy, efficiency, and patient care. Specifically, AI in GI pathology enhances diagnostic accuracy, streamlines workflows, provides predictive insights, integrates multimodal data, supports research, and aids in education and training, ultimately improving patient care and outcomes. This review summarized the latest developments in the role and scope of AI in AP with a focus on GI pathology. The main aim was to provide updates and create awareness among the pathology community.

Key Words: Gastrointestinal pathology; Digital pathology; Artificial intelligence; Machine learning; Deep learning; Precision diagnostics

Core Tip: Anatomic pathology remains largely subjective compared to other diagnostic laboratory fields. However, the digitization of tissue sections and the development of artificial intelligence-based technologies are rapidly advancing image-based diagnostics in anatomic pathology including gastrointestinal pathology. These technologies allow pathologists to make diagnoses more quickly and accurately, particularly for time-consuming and repetitive tasks, leading to higher volumes and faster turnaround times. Increasing awareness of the potential uses and benefits of these emerging technologies is essential for the pathology community.

INTRODUCTION

Anatomic pathology (AP), particularly histopathology, represents the ground truth of medicine, providing the final and definitive test on which crucial treatment decisions are based, especially for cancer (Figure 1). Despite its critical role, AP has remained an analog enterprise, using processes developed in the early 20^th century. Tissue preparation and diagnosis are still largely manual and subjective[1,2]. Diagnoses are based on the visualization and assessment of tissue sections on glass slides under a light microscope, making the process highly dependent on the pathologist’s interpretation.

Figure 1 Pathological diagnosis is the final and definitive test that informs all subsequent therapy decisions by clinicians, particularly in oncology.

The diagnostic process is a complex mental exercise requiring multitasking and the coordination of observation, interpretation, and integration of information. This process yields continuous variables that pathologists use to drive classification systems, which clinicians use to make major therapeutic decisions (Figure 1). While cost-effective, this process is prone to significant inter- and intra-pathologist variation and diagnostic errors and is often time-consuming and tedious. The integration of multiple ancillary diagnostic tests, such as immunohistochemistry and molecular assays, adds to the complexity and demands on pathologists. Moreover, the shortage of pathologists globally exacerbates these challenges, highlighting the need for precision diagnostics, particularly in cancer treatment[3-5].

Traditionally, AP has been slow to embrace digital technology, but this is steadily changing. Many pathology laboratories worldwide have partially or completely transitioned to digital pathology (DP) workflows[6-10]. Advanced artificial intelligence (AI) techniques, such as machine learning (ML) and deep learning (DL), are set to transform AP into a more objective, efficient, and transparent discipline[11,12]. However, many pathologists, particularly in developing countries, have limited knowledge of AI and its vast potential[13,14].

This review aimed to provide a simplified overview of the latest developments in the role and scope of AI in pathology, with a specific focus on gastrointestinal (GI) pathology. It explored how AI can be used to diagnose and predict diseases, highlighting its benefits for routine histopathology practice. The goal was to update pathologists and other healthcare providers about these emerging diagnostic technologies and raise awareness.

LITERATURE REVIEW

A comprehensive search was conducted across multiple databases, including PubMed, Google Scholar, Scopus, and Web of Science, covering publications from 2010 to 2024. Keywords included “Artificial Intelligence,” “Machine Learning,” “Deep Learning,” “Gastrointestinal Pathology,” “Diagnosis,” and “Histopathology.” All types of studies in the English language were selected based on relevance, focusing on AI applications in GI pathology. The full texts of these articles were carefully read to extract the relevant points for writing this review article.

DEFINITION OF AI AND ITS APPLICATIONS

AI is a field of computer science that enables computers to perform tasks that typically require human intelligence, such as learning, pattern recognition, planning, problem-solving, and reasoning[15-17]. AI relies heavily on data for its operations, and the digitization of glass slides in DP workflows provides vast amounts of pixel data for AI applications. AI can be considered a part of data science[18-21].

Initially, AI comprised simple “if-then” algorithms but has since advanced to include complex algorithms that perform tasks akin to the human brain. The advent of DL has expanded the capabilities of AI, allowing systems to analyze data and images using multiple layers and learn from big data. ML and DL are subsets of AI (Figure 2), with DL being a more advanced form capable of solving complex problems using neural networks (Figure 3)[22-24].

Figure 2 Relationship of artificial intelligence and its more advanced forms, machine learning and deep learning.

Figure 3 A simplified version of artificial neural network utilized in artificial intelligence algorithms.

AI represents a significant turning point for human society, comparable to the industrial revolution. It is a general purpose technology applicable in various fields, much like electricity. The role of AI in healthcare is growing rapidly, particularly in biomedical research and clinical practice. Modern techniques now generate vast amounts of data, which AI can analyze for patterns, enhancing diagnostics and treatment decisions[25-29]. AI can detect subtle pathological alterations, predict therapy responses, and improve workflow efficiency in pathology[30].

In the diagnostics arena, radiologists have been early adopters of AI for image processing and interpretation[31,32]. Pathologists, facing greater visual data complexity in microscopic images, have been slower to adopt AI. However, the shift towards DP and whole slide imaging (WSI) is laying the groundwork for computational pathology technologies. With the recent advancements in AI for computer vision, it is expected that AI will soon support pathologists in various DP tasks. Concurrently, significant progress in DL has created a synergy with AI, enabling image-based diagnostics within the DP context. Efforts are underway to develop AI tools that save pathologists time and reduce errors[33]. Integrating AI systems into AP practices will require fully digital imaging platforms, updating outdated information technology infrastructures, modifying laboratory and pathologist workflows, establishing appropriate reimbursement models, and ensuring pathologists’ active participation for buy-in and oversight. New regulations, designed to address the unique aspects and limitations of AI, are being developed to ensure its safe and effective use[34]. The recent Food and Drug Administration approval of WSI systems opens significant opportunities for AI-assisted pathological diagnosis, promising faster, more accurate, and cost-effective diagnostics[35-40].

ROLE OF AI IN ROUTINE AP DIAGNOSTIC WORKFLOW

AI is increasingly being integrated into AP to enhance diagnostic efficiency and accuracy, reduce turnaround times, and improve patient care[41,42]. AI algorithms can analyze DP images with high speed and accuracy, assisting pathologists in identifying and quantifying specific features such as cell structures, mitoses, tissue patterns, and abnormalities. This reduces subjectivity, minimizes diagnostic errors, and ensures consistent results[43-47].

AI-driven workflow management tools can streamline daily tasks, prioritize cases based on urgency, and help pathologists allocate their time effectively. AI can also integrate patient data, provide decision support tools, and assist in quality control and compliance with regulatory standards. AI tools can develop predictive models for disease outcomes and support research by analyzing vast datasets[48,49]. Although technical implementation has become less challenging, much work is needed to integrate AI into routine AP workflows. AI can also enhance the understanding of disease biology by analyzing DP images to identify patterns and features not visible to the human eye. This can aid in discovering new biomarkers and treatments, benefiting both diagnostics and research[50,51].

ROLE OF AI IN GI PATHOLOGY

AI technology is poised to revolutionize GI pathology, offering numerous current and potential applications (Figure 4). By processing digitized images of tissue samples, AI tools enhance the precise and effective identification of various GI disease processes, including inflammatory and neoplastic conditions such as colitis, Crohn’s disease, and colorectal cancer (CRC). The integration of AI in GI pathology significantly improves the precision, speed, and quality of diagnostic and therapeutic decision-making processes, ultimately benefiting patient care. Additionally, AI can standardize quality control in GI pathology, ensuring accurate and consistent results across samples[52-55].

Figure 4 Four main roles of artificial intelligence-based tools in gastrointestinal pathology. AI: Artificial intelligence; GI: Gastrointestinal.

AI has been extensively studied for endoscopic diagnosis of GI tract disorders, demonstrating significant promise. It is expected that AI will primarily assist endoscopists with tasks such as detection, characterization, and segmentation. AI has the potential to enhance colonoscopy-based colorectal screening and monitoring by reducing unnecessary expenses and improving quality. Real-time computer-assisted polyp identification can enhance screening and monitoring quality, as measured by adenoma detection rates. Optical biopsies using computer-assisted diagnosis can identify low-risk polyps, supporting resect-and-discard or diagnose-and-leave strategies, thereby reducing unnecessary costs. Recent meta-analyses indicated that AI tools significantly increased colorectal neoplasia detection, regardless of initial adenoma features[56-58]. Furthermore, AI is useful in identifying upper GI pathological processes, including both neoplastic and non-neoplastic lesions[59,60].

In the GI tract, precancerous lesions and invasive tumors are routinely biopsied or excised for histopathological workup. Early and accurate diagnosis is a primary responsibility of pathologists, and AI can assist in achieving this objective. Numerous reports have documented AI-assisted diagnosis of both neoplastic and non-neoplastic GI diseases. For instance, Korbar et al[61] trained a model to distinguish between five prevalent types of colorectal polyps with an overall accuracy of 93% using a dataset of over 400 WSIs. Wei et al[62] demonstrated that neural networks trained to identify colorectal polyps on WSIs from one institution performed similarly to local pathologists when applied to WSIs from other institutions. Efforts have also been made to automate the diagnosis of preneoplastic and neoplastic lesions, such as Barrett’s esophagus or gastric adenomas/adenocarcinomas[60].

AI models also show promise in predicting therapy response or prognosis from WSI analysis. Among all cancer types, GI and liver tumors have notably driven computational oncology forward. AI can extract complex information from digital images of GI and hepatic malignancies, providing clinical, biological, and molecular insights that are not accessible to the naked eye. By identifying the most predictive tissue areas, AI reduces the cognitive burden on pathologists, enhancing their efficacy in histopathological characterization and risk assessments of GI preneoplastic and neoplastic lesions. In biliary tract cancer, DL can identify tissue features predictive of clinical outcomes. DP images and tissue microarrays from CRC have shown the efficacy of DL in prognostic prediction across all tumor stages. The histomorphology of gastric cancer (GC) is more complex and variable than CRC, leading to fewer investigations using DL for GC. Most of this research has focused on tumor detection rather than prognostication[63-66].

Routine processing of surgical and biopsy specimens from various GI tract tumors involves investigating molecular biomarkers that predict responses to targeted therapy. Specific genetic events in GI and hepatobiliary cancers are linked to morphological features identified in hematoxylin and eosin sections. AI-based algorithms on WSIs have been successfully used as surrogate markers for these alterations[66-69]. For example, CRC serves as a model disease due to the abundance of pathology samples. Identifying microsatellite instability (MSI) is crucial because immune-modulating treatments significantly affect MSI tumors. MSI identification has major implications for patients and their families, necessitating further investigation to identify Lynch syndrome. Although immunohistochemistry techniques are typically used to identify MSI, not all patients are routinely screened. A study by Echle et al[70] examined 8836 CRC cases across all stages, developing an AI model that could identify MSI tumors from hematoxylin and eosin sections, maintaining performance even in biopsy samples with limited tissue and varying preprocessing methods. Other efforts have created models that accurately predict gene alterations from WSIs of hepatocellular carcinoma (HCC), GC, and other conditions.

AI-based pathology can predict gene expression and RNA sequencing data, holding great promise for clinical application. Developing DL models for prognostication that integrate clinical, biological, and genetic data is a promising approach. For example, Chaudhary et al[71] used RNA sequencing, microRNA sequencing, and methylation data to create a DL model predicting survival in HCC patients, demonstrating its efficacy across different HCC patient cohorts.

AI ALGORITHMS IN GI PATHOLOGY

AI algorithms, particularly those based on ML and DL, have shown substantial potential in analyzing complex pathological data and are central to the advancements in GI pathology (Table 1). ML algorithms such as support vector machines, random forests, and k-nearest neighbors have been employed to classify and predict various GI conditions. These include supervised, unsupervised, and reinforcement learning algorithms (Figure 5). Supervised learning algorithms, such as support vector machines and random forests, are widely used for classification tasks. These algorithms require extensive feature engineering and domain expertise to identify relevant features from pathology images and clinical data. They are relatively simpler to implement and interpret compared to DL algorithms. They are also effective for structured data analysis and smaller datasets and feature faster training times and lower computational requirements. The need for manual feature extraction limits their use, and ML algorithms may not perform as well as DL in recognizing complex patterns in unstructured data like histopathological images[72-75].

Figure 5 Main artificial intelligence-based algorithms for use in gastrointestinal pathology.

Table 1 Main artificial intelligence-based algorithms for use in the gastrointestinal pathology field.

AI algorithms	Role in gastrointestinal pathology	Key uses	Examples
Machine learning	Assisting in diagnosis and classification of gastrointestinal diseases	Improved diagnostic accuracy, personalized treatment plans	Predictive models for colorectal cancer risk, classification of polyps in colonoscopy images
Deep learning	Analyzing endoscopic and histopathologic images	Enhanced image recognition, reduced human error	Convolutional neural networks for detecting and classifying lesions in endoscopic images
Natural language processing	Extracting relevant information from medical records and literature	Efficient data mining, real-time clinical decision support	Automated extraction of patient data from electronic health records for research and clinical use
Computer vision	Real-time analysis of endoscopic videos	Immediate feedback during procedures, increased detection rates of abnormalities	Detection of bleeding, polyps, and other abnormalities during live endoscopy procedures
Reinforcement learning	Optimizing treatment plans and clinical pathways	Adaptive learning from real-world outcomes, improved clinical decision-making	Personalized treatment strategies for inflammatory bowel disease based on patient response
Predictive analytics	Forecasting disease progression and patient outcomes	Proactive patient management, early intervention	Predicting flare-ups in Crohn’s disease, forecasting outcomes after gastrointestinal surgeries
Robotics integration	Enhancing precision in minimally invasive surgeries	Increased surgical precision, reduced recovery time	AI-assisted robotic surgery for gastrointestinal procedures, such as robotic-assisted colectomy
Genomic data analysis	Identifying genetic markers associated with gastrointestinal diseases	Personalized medicine, targeted therapies	Analyzing genetic data to find markers for conditions like Lynch syndrome and hereditary pancreatitis

AI: Artificial intelligence.

DL models, particularly convolutional neural networks, have revolutionized image analysis in GI pathology[76,77]. Convolutional neural networks can automatically learn hierarchical features from raw images, making them highly effective for tasks such as tumor detection and classification. They possess superior performance in image recognition tasks and are able to handle large and complex datasets. Automated feature extraction reduces the need for domain-specific knowledge. However, it requires substantial computational resources and large annotated datasets. It is difficult to interpret and explain the decision-making process (black-box nature). Recurrent neural networks, including their variants like long short-term memory networks, are used for sequential data analysis. They are particularly useful in analyzing time-series data from endoscopic videos to detect abnormalities[78-84]. While ML algorithms are generally more interpretable, DL algorithms often provide higher accuracy due to their ability to learn complex patterns from large datasets. However, DL models require substantial computational resources and large labeled datasets, which can be a limitation[73,75,79,85].

DATA SOURCES IN GI PATHOLOGY

The performance of AI algorithms heavily depends on the quality and diversity of data sources. Common data sources in GI pathology include: (1) Histopathological images in the form of WSIs and tissue microarrays as the primary data sources; (2) Clinical data, such as electronic health records, patient demographics, clinical history, and endoscopy images; and (3) Publicly available datasets such as The Cancer Genome Atlas and Gastrointestinal Image Data Collection. Each of these sources has merits and demerits. Integration of clinical data, including patient demographics, medical history, and laboratory results, enhances the contextual understanding of GI pathology and improves the predictive power of AI models. The availability of large, well-annotated datasets is a significant challenge. Variability in image quality and staining techniques and differences in pathological practices across institutions can affect the generalizability of AI models. Additionally, integrating clinical data requires sophisticated data management systems to handle patient privacy and data security concerns.

PERFORMANCE METRICS OF AI ALGORITHMS

Sensitivity measures the ability of an AI algorithm to correctly identify positive cases (e.g., detecting cancerous lesions). Specificity measures the ability of an AI algorithm to correctly identify negative cases (e.g., ruling out benign conditions). Achieving high sensitivity and specificity is challenging due to the inherent variability in pathological samples. DL models often outperform traditional ML models in sensitivity and specificity due to their ability to learn intricate features from large datasets. However, there is a trade-off; models with high sensitivity might produce more false positives, reducing specificity. Balancing these metrics is essential to avoid misdiagnoses and unnecessary treatments[74,76,81,82].

CHALLENGES AND FUTURE PROSPECTS

AI is still in its early stages, and many pathology laboratories worldwide have yet to transition to a digital workflow to fully benefit from AI technologies. There are numerous obstacles to the widespread implementation of AI solutions in routine clinical practice, even in developed countries. Bringing an AI solution for pathology to market poses significant technological, business, and regulatory challenges. Although some clinical applications exist, the overall introduction of AI into medical practice has been slow and not without ethical concerns[86-90].

Despite significant research developments in AI-based techniques in recent years, only a few AI solutions have become commercial products for routine use. Consequently, much of the potential of AI remains untapped. Research models need further development, improvement, and integration into the information technology infrastructure of clinical laboratories before they can be used in routine pathology workflows. Additionally, commercial success requires a profitable business model in most countries, and pathologists need to be reimbursed for using the product. AI solutions are also classified as medical devices and thus require regulatory approval before they can be sold as products[91-94].

CONCLUSION

The role and scope of AI are expanding in GI pathology, with the potential to improve diagnostic accuracy, efficiency, and patient care. Increasing awareness among the pathology community about these emerging technologies is essential to realize their full potential and revolutionize diagnostics, prognostics, and theranostics in GI pathology.