Current Translational Medicine, ISSN NoIssnPrint print, 0000-0000 online, Open Access
Article copyright, the authors; Journal compilation copyright, Curr Transl Med and Elmer Press Inc
Journal website https://ctm.elmerpub.com

Review

Volume 1, 2026, e8

Artificial Intelligence in Functional Lumen Imaging Probe Panometry: Toward Automated Functional and Biomechanical Endoscopy

Figure

↓  Figure 1. Conceptual pipeline for artificial intelligence applied to functional lumen imaging probe (FLIP) panometry. Volume-controlled balloon distension during sedated endoscopy generates paired cross-sectional-area and intra-balloon-pressure data (impedance planimetry), which are rendered as a FLIP panometry topographic plot of esophagogastric junction opening and the distension-induced contractile response. Computational models (supervised classifiers, deep neural networks, unsupervised representation learning, and physics-informed/biomechanical models) convert these inputs into clinical outputs: standardized motility classification, probabilistic diagnosis, real-time intraprocedural decision support, biomechanical phenotyping, and prediction of treatment outcomes. The heatmap is an illustrative schematic, not patient data. CNN: convolutional neural network; ML: machine learning.
Figure 1.

Tables

↓  Table 1. Common FLIP Panometry Metrics and Their Clinical Interpretation
 
MetricWhat it measuresClinical interpretation and main useKey reference(s)
C2D2: Compliance-Contractile response-Distensibility plateau-Diameter composite score; DI: distensibility index; DP: distensibility plateau; EGJ: esophagogastric junction; FLIP: functional lumen imaging probe.
DINarrowest cross-sectional area divided by intra-balloon pressureCore measure of how readily a sphincter or segment opens under load; reduced values indicate impaired opening[4, 5]
EGJ-DIDistensibility index measured at the esophagogastric junctionReduced EGJ-DI indicates impaired EGJ opening (e.g., achalasia, EGJ outflow obstruction); validated thresholds distinguish obstruction[8]
Maximum EGJ diameterLargest EGJ diameter achieved during volumetric distensionComplements EGJ-DI in grading junction opening (normal/inconclusive/reduced)[5, 9]
DPPlateau of esophageal body diameter reached at higher fill volumesMarker of esophageal body distensibility; reduced in fibrostenotic eosinophilic esophagitis and linked to stricture and impaction risk[10]
Esophageal body complianceChange in esophageal body volume per change in pressureCaptures wall remodeling beyond the distensibility plateau; reduced compliance reflects a stiffer, remodeled esophagus[10]
Contractile response patternPresence, vigor, and direction of distension-induced (secondary) contractionsClassified as normal, diminished, disordered, absent, or spastic; defines the motor component of panometry[9]
Composite physiomechanical score (e.g., C2D2)Integrated score combining compliance, contractile response, distensibility plateau, and EGJ diameterSummarizes esophageal physiomechanical function (e.g., in eosinophilic esophagitis) and has been associated with treatment response[11]

 

↓  Table 2. Representative Studies Applying Artificial Intelligence or Computational Modeling to FLIP/Impedance-Planimetry Data (Part A, direct evidence) and to Closely Related Esophageal Physiology Assessed With Other Modalities (Part B, Indirect Evidence)
 
Study (Ref)Setting and dataMethod and taskSampleValidation typeReference standardKey reported performanceMain limitation
Validation type and reference standard are reported as described by the original authors; “internal” denotes training and testing within the same dataset, without evaluation in an independently withheld center or cohort. AUC: area under the receiver-operating-characteristic curve; CNN: convolutional neural network; ConvLSTM: convolutional long short-term memory; EGJ: esophagogastric junction; EU: Europe; FLIP: functional lumen imaging probe; HRM: high-resolution manometry; LES: lower esophageal sphincter; ML: machine learning; MRI: magnetic resonance imaging; PPV: positive predictive value; US: United States; v2.0: version 2.0; v4.0: version 4.0; xGB: extreme gradient boosting.
Part A. Studies using functional lumen imaging probe or impedance-planimetry data (direct evidence)
  Carlson et al [20], 2021Single center; FLIP panometry; retrospectiveSupervised decision tree and random forest on engineered contractility and pressure features; identify achalasia subtypes180 patientsInternal; training and held-out test split of the same cohortManometric achalasia subtype (Chicago Classification)Type III vs non-spastic 90% (train)/78% (test) accuracySmall single-center cohort; discrimination among individual subtypes limited; dependent on hand-engineered features; no external validation and no outcome endpoint
  Schauer et al [24], 2022Single center; FLIP metrics; retrospectiveBayesian additive regression trees; estimate probability of EGJ obstruction557 patients + 35 controlsInternal; leave-one-out evaluation with a holdout test set, both from the same cohortExpert-assigned diagnosis of EGJ obstruction based on HRM and complementary clinical testingAccuracy 89–90%; AUC 0.95–0.97Discrimination reported without calibration; single center; expert-defined reference rather than treatment response; no external validation
  Halder et al [25], 2022Single center; FLIP panometry processed through a one-dimensional inverse mechanical modelVariational autoencoder over mechanics-based parameters; unsupervised mapping of esophageal disorders onto a virtual disease landscapeCohort spanning motility disorders, eosinophilic esophagitis, reflux disease, and scleroderma esophagusInternal; latent-space clustering compared with established diagnostic categoriesExisting clinical diagnostic categories (model itself unsupervised)Disorders occupied distinguishable regions of the landscape; disease progression over time and treatment effect displayed for individual patientsProof of concept; no independent clinical endpoint; treatment-effect demonstration illustrative
  Kou et al [21], 2023Single center; raw FLIP output rendered as panometry heatmapsConvolutional neural network (deep learning); classify esophageal motility678 patients + 35 controlsInternal; held-out evaluation within the same cohortExpert panometry interpretation, with HRM for achalasia89% accuracy in each binary stage vs experts; 93% of 28 HRM-confirmed achalasia test cases identifiedAgreement with expert labels rather than clinical outcome; single center and single probe platform; limited interpretability; no external validation
  Halder et al [27], 2024Single center; FLIP panometry; retrospectiveMechanics-based parameters, variational autoencoder, and linear discriminant analysis; assign diagnostic probabilities on the virtual disease landscape740 subjectsInternal; comparison with FLIP and HRM diagnoses in the same cohortExpert FLIP and HRM (Chicago Classification) diagnosesSeparation of normal EGJ opening from EGJ obstruction aligned closely with both FLIP and HRM diagnosesSingle center; expert-label target; no withheld-center or outcome validation
  Fass et al [22], 2025Single center; FLIP panometry heatmaps; retrospectiveTransformer deep-learning classifier trained to panometry classification v2.0; classify motility as normal, “achalasia” (including conclusive EGJ outflow obstruction), or abnormal985 patients (719 training, 127 validation, 139 test)Internal; patient-level training, validation, and test split without patient overlapBlinded expert diagnosis from HRM (Chicago Classification v4.0), with record review and timed barium esophagram where inconclusive84% accuracy on the test set versus 71% for the earlier convolutional model; PPV 91% for normal motility and 84% for achalasiaResearch letter with limited methodological detail; single center; single platform
  Halder et al [28], 2025Single center; FLIP panometry; software frameworkWeb framework combining clinical metrics, mechanics-based metrics, and virtual-disease-landscape probabilities121 subjectsInternal; framework output compared with expert categorizationExpert clinician categorization of EGJ openingFramework metrics matched expert diagnosis; misclassifications few and mainly between adjacent categoriesSmall validation set; single center; expert-label target; tool validation rather than clinical evaluation
  Mascarenhas et al [23], 2026Multicenter (EU + US); FLIP panometryMultiple ML classifiers (AdaBoost, random forest, gradient boosting); classify panometry per Dallas Consensus105 exams, five centersInternal; patient-level training and testing split within the pooled multicenter sample (not center-level external validation)Expert Dallas Consensus classificationAccuracy 84.9–86.7%; AUC 0.89–0.97 (Results/Table 1 of ref. [23])Small number of examinations; no withheld center for external testing; single commercial platform; expert labels rather than outcomes
Part B. Related studies using other modalities (indirect evidence)
  Kou et al [26], 2021Single center; raw HRM swallowsVariational autoencoder (unsupervised); learn motility representations> 32,000 swallowsInternal; reconstruction and latent-space assessmentNone (unsupervised; latent features compared with established manometric metrics)Latent features encoded contraction strength, propagation, relaxationHRM rather than FLIP; no diagnostic or clinical validation; latent features not linked to management decisions
  Halder et al [29], 2023Mechanics-informed MRI of the esophagusPhysics-informed neural network; estimate wall stiffness and active relaxation1 healthy volunteerInternal; physical-consistency and simulation checksModel-based mechanical quantities and MRI-derived measurementsRecovered esophageal mechanical properties and missing LES dataImaging modality other than FLIP; feasibility stage; very small sample; no clinical endpoint
  Wang et al [30], 2021HRM dataset (context)3D CNN + bidirectional ConvLSTM; classify esophageal motility over timeHRM dataset, single sourceInternal validationExpert HRM classification91.3% accuracy; 90.5% sensitivity; 95.9% specificityHRM rather than FLIP; single data source; expert labels; no external or outcome validation
  Mascarenhas et al [31], 2025Multicenter, multidevice HRM (context)Gradient boosting/xGB; detect Chicago Classification disorders618 exams, three centersInternal cross-validation across pooled multidevice dataChicago Classification v4.0 expert labelsEGJ outflow accuracy 0.942; AUC 0.921HRM rather than FLIP; internal rather than held-out-center validation; expert labels rather than outcomes