Scope and caveats: Artificial intelligence in gastroenterology

Abstract

The use of Artificial intelligence (AI) has evolved from its mid-20^th century origins to playing a pivotal tool in modern medicine. It leverages digital data and computational hardware for diverse applications, including diagnosis, prognosis, and treatment responses in gastrointestinal and hepatic conditions. AI has had an impact in diagnostic techniques, particularly endoscopy, ultrasound, and histopathology. AI encompasses machine learning, natural language processing, and robotics, with machine learning being central. This involves sophisticated algorithms capable of managing complex datasets, far surpassing traditional statistical methods. These algorithms, both supervised and unsupervised, are integral for interpreting large datasets. In liver diseases, AI's non-invasive diagnostic applications, particularly in non-alcoholic fatty liver disease, and its role in characterizing hepatic lesions is promising. AI aids in distinguishing between normal and cirrhotic livers and improves the accuracy of lesion characterization and prognostication of hepatocellular carcinoma. AI enhances lesion identification during endoscopy, showing potential in the diagnosis and management of early-stage esophageal carcinoma. In peptic ulcer disease, AI technologies influence patient management strategies. AI is useful in colonoscopy, particularly in detecting smaller colonic polyps. However, its applicability in non-academic settings requires further validation. Addressing these issues is vital for harnessing the potential of AI. In conclusion, while AI offers transformative possibilities in gastroenterology, careful integration and balancing of technical possibilities with ethical and practical application, is essential for optimal use.

Key Words: Machine learning, Neural networks, Diagnosis, Work-flow, Ethics, Image, Polyps, Hepatoma

Core Tip: Artificial intelligence helps in the early identification, management and prognostication of gastrointestinal diseases through applications in endoscopy and histopathological interpretation. Proof of concept studies exist for all of these, but need validation by randomized clinical trials before they can be incorporated into clinical work flow.

INTRODUCTION

Artificial intelligence (AI) has origins dating back to the mid-20^th century[1]. Its application, particularly in gastroenterology, has been more recent, being dependent on the availability of digital data and powerful computational hardware. AI is now used in diagnosing and predicting treatment responses for a spectrum of gastrointestinal and hepatic disorders[2-5].

AI encompasses multiple fields and is often defined as a computer's ability to perform tasks requiring human-like cognition[5], or as a machine that simulates complex human thinking[2].

In contrast to traditional statistical methods, AI processes data using a large number of variables and sophisticated formulas, making it possible to perform otherwise impractical analyses[2].

NOMENCLATURE AND CLASSIFICATION OF AI

AI broadly includes three areas: (1) Machine learning, which involves artificial neural networks, deep learning, and convolutional neural networks; (2) Natural language processing; and (3) Robotics. Machine learning techniques are principally used in gastroenterology[6].

Machine learning can be considered an advanced statistical approach to uncover relationships among various parameters; it utilizes algorithms such as linear regression for predicting relationships between variables, and classification algorithms like support vector machines and random forests for categorizing data.

Neural networks are more complex, utilizing nodes to determine calculated parameters, thereby allowing the use of intricate formulas. Deep learning networks, which are multi-layered, enable more advanced learning, are used in image processing[2].

Machine learning models are categorized into supervised and unsupervised. Supervised models label each sample, while unsupervised models aim to discover data structures without labels.

APPLICATIONS OF AI IN GASTROENTEROLOGY

AI's applications in gastroenterology promise to enhance patient care by reducing diagnostic errors[6]. They are employed in various conditions including gastritis, gastrointestinal bleeding, gastric malignancy, non-alcoholic fatty liver disease (NAFLD), cirrhosis, inflammatory bowel disease, colorectal polyps and cancer, and computer-aided endoscopy. Other potential applications include Helicobacter pylori infection, celiac disease, and pancreatic lesions[5]. A MEDLINE database search indicates that China and the USA are leading in AI research in this field[3].

LIVER DISEASES

In liver diseases, AI applications span from hepatocellular carcinoma to NAFLD, benign tumours, and viral hepatitis.

Non-invasive methods like ultrasound or transient elastography are used to identify NAFLD, now classified as metabolic dysfunction-associated steatohepatitis[7]. Probabilistic neural networks differentiate normal livers from those with cirrhosis in NAFLD patients, with the gold standard being liver biopsy. The area under the curve for this method ranges between 0.857 and 0.901[8]. AI also aids in automating histopathological examination, achieving high accuracy in characterizing alterations found in NAFLD[9]. Predictive models using multiple data sources, including electronic medical records, imaging, and biomarkers, have improved accuracy in identifying at-risk patients[10,11].

Hepatic mass lesions can be interpreted with high accuracy by the use of AI. Deep learning methods achieved receiver operating curves of 0.93 in lesion differentiation and 0.916 in characterization[12]. AI also aids in prognosticating established hepatocellular carcinoma[11] and predicting graft failure following liver transplantation[4].

However, when compared with standard scores, AI did not significantly improve the accuracy of short-term predictions for readmission and mortality risks in patients with cirrhosis[13].

ESOPHAGEAL LESIONS

AI methodologies improved the identification of suspicious lesions during upper GI endoscopy; this was reported in the differentiation of dysplasia and early neoplastic changes in Barrett's esophagus[14]. Other technologies such as white-light endoscopy/narrow band imaging, wide-area transepithelial sampling, and volumetric laser endomicroscopy lend themselves to machine learning[14]. As of 2020, AI algorithms were shown to be effective in diagnosing and thereby improving the outcomes of early-stage esophageal carcinoma.

ACID PEPTIC DISORDERS

In peptic ulcer disease, AI is useful in diagnosis, management, and complications[15]. Helicobacter pylori, a significant pathogenic factor, can be identified using AI. The first application was reported in 2004[16]; recent studies employ convoluted neural network models on large datasets, achieving high sensitivity, specificity, and accuracy[17].

AI also assists in diagnosing and differentiating peptic ulcers with wireless capsule endoscopic images by the use of deep learning[18,19]. It achieved an overall sensitivity of 89.7%. AI can also help identify infections, ulcers, polyps, and submucosal xanthomas[20,21].

For rare cases requiring surgical intervention, AI finds application in robot-assisted minimally invasive surgery, and in predicting complications like bleeding and perforation by the use of data from electronic medical records[15,22].

COLONOSCOPY

AI aids in classifying structures and has notably improved polyp detection rates, particularly in identifying polyps less than 5cm in size[2]; these are often missed by conventional procedures.

AI is also effective in grading remission of ulcerative colitis[23,24] and in assessing video capsule endoscopy for ulcers and bleeding detection, which is a time-consuming task[25].

However, mixed results were reported for polyp detection using computer-aided endoscopy in non-academic community-based practice[26], indicating the need for further studies[27].

PROS AND CONS OF AI IN GASTROENTEROLOGY

The potential of AI to enhance quality of care is significant, but integration into clinical workflows remains a challenge[28]. In specific tasks, AI-based devices match or even surpass expert gastroenterologists in identifying and differentiating neoplasms in the gastrointestinal tract. However, for adoption into routine clinical practice, randomized trial validation is necessary[28].

Other factors such as disease prevalence, physician competence, and human-machine interaction also affect AI's clinical benefit[29]. Despite the availability of commercial AI tools in the USA and Europe, their integration into clinical workflows is still a work in progress before the full potential is realized[30].

ETHICAL AND LEGAL ASPECTS

The growing use of AI in clinical practice brings ethical and legal challenges such as data privacy and security, method reliability and safety, and ensuring fairness, inclusivity, transparency, and accountability[31]. The extent of reliance on AI for decision-making is a key consideration[28], given that the ultimate responsibility rests on the end user, viz the physician.

AI has the potential for bias in clinical problem selection, variable choices, algorithm development, and system use[32]. In gastroenterology, as in other fields, training sets must be inclusive and diverse to avoid bias in diagnosing diseases with varying prevalence rates.

To mitigate potential biases, health equity goals should be incorporated early in algorithm development by involving technically diverse research teams. Regulatory standards should include pre-deployment audits to ensure algorithmic performance equality[32].

CONCLUSION

AI in gastroenterology has primarily been applied in endoscopic image analysis, radiology, and histopathology to aid in early detection of lesions and appropriate treatment. While its role in diagnostic endoscopy is expanding, evidence for improved clinical outcomes in real-life scenarios remains to be established.

Issues surrounding human-machine interaction, AI integration into clinical culture and practice, and the balance between AI-assisted management and practitioner skill maintenance need to be addressed[33]. Generative AI, such as ChatGPT, launched in late November 2022[34], has become pervasive in medicine, including gastroenterology. While beneficial in diagnosis in complex scenarios[35], privacy and legal concerns arise, especially when scientific publications could eventually heavily on AI-generated results[36].

ACKNOWLEDGEMENTS

We thank Mr Venkat Yarabati for assistance in the preparation of this manuscript.