Harnessing artificial intelligence in gastroenterology and hepatology: Current applications and future perspectives

Table 1 Comparison of key artificial intelligence approaches relevant to gastroenterology and hepatology

AI technique	Input data type	Interpretability	Computational cost	Common clinical applications	Limitations
Logistic regression	Structured tabular data	High	Low	Risk scoring (e.g., cirrhosis, NAFLD), binary classification	Assumes linear relationships, limited complexity handling
Decision trees/random forests	Structured data, some semi-structured	Moderate to high	Moderate	Prognostic models (e.g., HCC recurrence), treatment stratification	Can overfit, less effective on unstructured data
SVM	Structured data, imaging (preprocessed)	Low to moderate	Moderate	Classification tasks (e.g., benign vs malignant lesions)	Less scalable, needs careful kernel tuning
CNNs	Imaging data (e.g., endoscopy, CT, MRI)	Low	High	Polyp detection, liver lesion classification, fibrosis staging	Black-box nature, high data requirements
RNNs/LSTMs	Time-series data, text sequences	Low	High	Monitoring biomarkers over time, EHR text analysis	Difficult to train, prone to vanishing gradients
Transformers/LLMs	Natural language, multimodal inputs	Moderate to low	Very high	Summarization of clinical notes, patient stratification via EHR	Expensive to fine-tune, interpretability challenges
Autoencoders/unsupervised learning	Imaging, high-dimensional data	Low	Moderate to high	Anomaly detection, feature extraction	Requires careful architecture design, lacks direct supervision
Federated learning	Decentralized structured/unstructured data	Moderate	High	Multi-center model training without data sharing	Complex orchestration, risk of data heterogeneity bias

AI: Artificial intelligence; SVM: Support vector machines; CNNs: Convolutional neural networks; RNNs: Recurrent neural networks; LLMs: Large language models; NAFLD: Non-alcoholic fatty liver disease; HCC: Hepatocellular carcinoma; CT: Computed tomography; MRI: Magnetic resonance imaging; LSTMs: Long short-term memory networks; EHR: Electronic health record.

Table 2 Illustrative examples of artificial intelligence applications across different stages of the patient journey in gastroenterology and hepatology

Stage of patient journey	Clinical context	Representative AI applications	Benefits/added value	Hype cycle stage	Ref.
Risk stratification and screening	Asymptomatic or high-risk individuals	Predictive modeling for NAFLD, HCC, CRC risk	Early identification of at-risk patients, targeted screening programs	Slope of enlightenment	[1]
		Polygenic/biomarker-based risk stratification using EHR and genomics
Diagnosis	Symptomatic presentation or incidental findings	AI-assisted polyp detection during colonoscopy (real-time CADe)	Increased diagnostic accuracy, real-time decision support, reduced miss rates of small/flat lesions	Plateau of productivity (for CADe in CRC); peak of inflated expectations (for capsule endoscopy CNNs)	[4,12,15]
		Image-based classification of liver lesions (CNNs)
		Capsule endoscopy with U-Net architectures
Staging and prognostication	Confirmed disease (IBD, cirrhosis, cancer)	Fibrosis staging via elastography DL	Improved risk assessment, personalized follow-up plans	Slope of enlightenment	[4,15]
		AI HCC recurrence risk prediction (random survival forests)
		Prognostic models (ML, MELD + AI)
Treatment planning	Therapeutic decision-making	AI-augmented MDT support for IBD biologics	Data-informed, individualized therapeutic pathways	Innovation trigger → early peak	Radiomics-based TACE prediction, AUC 0.78-0.85[36]
		RL models for drug sequencing
		Radiomics + ML for TACE suitability in HCC
Therapy monitoring	During pharmacologic, endoscopic, or surgical therapy	AI-based monitoring of treatment response (e.g., colectomy trends)	Dynamic tracking, early alerts, adaptive therapy modulation	Peak of inflated expectations	Colectomy prediction, AUROC 0.80-0.83[36]; NLP AE detection, recall 074-0.82[41,43]
		NLP for adverse event detection
Follow-up and surveillance	Post-therapy or remission phase	Predictive models for relapse in IBD	Enhanced vigilance, resource optimization, reduced recurrence risk	Slope of enlightenment	IBD relapse models, AUROC 0.79-0.82[20,22]
		Surveillance of HCC post-resection using ML
Patient engagement and education	Across all stages	AI chatbots for symptom triage	Empowered patients, improved adherence, scalable support	Peak of inflated expectations (for chatbots); innovation trigger (for advanced NLP coaching)	[34]
		Personalized education via NLP-based tools
		Digital coaching for diet/lifestyle adherence

AI: Artificial intelligence; NAFLD: Non-alcoholic fatty liver disease; HCC: Hepatocellular carcinoma; CRC: Colorectal cancer; CADe: Computer-aided detection; CNNs: Convolutional neural networks; IBD: Inflammatory bowel disease; DL: Deep learning; ML: Machine learning; MELD: Model for end-stage liver disease; MDT: Multidisciplinary team; TACE: Transarterial chemoembolization; AUC: Area under the curve; NLP: Natural language processing; AUROC: Area under the receiver operating characteristic curve; AE: Adverse event.