Present and future of endoscopy precision for inflammatory bowel disease

Abstract

Several advanced imaging techniques are now available for endoscopists managing inflammatory bowel disease (IBD) patients. These tools, including dye‐based and virtual chromoendoscopy, probe‐based confocal laser endomicroscopy and endocytoscopy, are increasingly innovative applications in clinical practice. They allow for a more in‐depth and refined evaluation of the mucosal and vascular bowel surface, getting closer to histology. They have demonstrated a remarkable ability in assessing intestinal inflammation, histologic remission, and predicting relapse and favorable long‐term outcomes. In addition, the future application of molecular endoscopy to predict biological drug responses has yielded preliminary but encouraging results. Furthermore, these techniques are crucial in detecting and characterizing IBD‐related dysplasia, assisting endoscopic mucosal resection and submucosal dissection towards a surgery‐sparing approach. Artificial intelligence (AI) holds great potential in this promising landscape, as it can provide an objective and reproducible assessment of inflammation and dysplasia. Moreover, it can improve the prediction of outcomes and aid in subsequent therapeutic decision‐making. This review aims to summarize the promising role of state‐of‐the‐art advanced endoscopic techniques and related AI‐enabled models for managing IBD, paving the way for precision medicine.

INTRODUCTION

The field of endoscopic tools available is continuously and rapidly evolving. Their contribution to the diagnosis, outcomes prediction, and therapeutic management of inflammatory bowel disease (IBD) and colitis‐related cancer is becoming increasingly valuable for personalized medicine. ¹

These new instruments, already part of the endoscopists' armamentarium, enable a more accurate assessment of inflammation and dysplasia, leading to improved characterization of mucosal healing (MH) by bringing endoscopy closer to histology for defining histologic remission (HR). ^{2 , 3}

Chromoendoscopy is an advanced tool that enhances intestinal mucosa and vascular pattern evaluation. ⁴ Dye‐based chromoendoscopy (DCE) involves staining the colon with various agents, such as methylene blue and indigo carmine, providing detailed mucosal characterization. Conversely, virtual electronic chromoendoscopy (VCE) is a dye‐less technology that exploits different light filters or postprocessing algorithms to highlight surface and vessel architecture. Various VCE platforms are currently available, including narrow‐band imaging (NBI) (Olympus, Tokyo, Japan), iSCAN‐optical enhancement (OE) (Pentax, Tokyo, Japan), and blue‐light imaging (BLI)/linked color imaging (LCI) (Fujifilm, Tokyo, Japan). Additionally, novel endoscopic imaging techniques enable in vivo evaluation of microscopic features of the colonic mucosa as a real‐time histology. ⁵ For instance, probe‐based confocal laser endomicroscopy (pCLE) provides imaging of the mucosa at the cellular and subcellular levels using intravenous fluorescent agents, resulting in highly magnified images with up to 1000‐fold magnification. Also, the endocytoscope (Olympus) enables detailed real‐time examination of the mucosal surface with up to 1400‐fold magnification, rivaling histology in evaluating crypt architecture, cellular infiltration, and micro‐vessel impairment.

Considering the remarkable ability of these techniques to predict endoscopic and histologic healing, a growing interest is arising in their potential for predicting disease flares and clinical outcomes, such as hospitalization, need for surgery, and changes in advanced therapies.

Great expectation is also placed on artificial intelligence (AI)‐assisted endoscopy. The standardized and improved assessment of inflammation and dysplasia through AI could revolutionize clinical and therapeutic practices in IBD, with relevant benefits for personalized patient management, clinical trials, and cost‐effectiveness. ⁶ Finally, advanced endoscopy for therapeutic purposes shows promise, but further studies are needed to evaluate its usefulness in clinical practice.

This narrative review aims to provide an overview of the innovative endoscopic tools available to clinicians and their potential to support precision medicine in IBD.

PRECISION ENDOSCOPY IN DIAGNOSIS AND ASSESSMENT OF INFLAMMATION AND DYSPLASIA IN IBD

Image‐enhanced endoscopy to assess inflammation

Endoscopic remission is a desirable long‐term target in both ulcerative colitis (UC) and Crohn's disease (CD). ³ Nonetheless, patients with endoscopic remission in current practice may experience disease relapse. ⁷ Having carefully excluded other contributing factors, such as lower treatment adherence, drug loss of response, and superimposed infections, the relapse may be attributed to a subtle histologic inflammation undetectable by standard endoscopy. ⁸ Hence, advanced endoscopic techniques, increasing image resolution, and allowing an in vivo and real‐time histologic evaluation of the mucosa can provide a finer characterization of the inflammation.

Dye‐chromoendoscopy

Only a few studies evaluated the ability of DCE to assess IBD extension and severity, since most of the work is focused on its use in identifying and evaluating dysplastic areas. ^{9 , 10} Nonetheless, Ibarra‐Palomino et al. ¹¹ investigated the usefulness of DCE in the determination of the extension and severity of UC. They found that DCE increased detection of areas with minimal inflammatory activity, enhancing endoscopic‐pathologic agreement for the assessment of UC severity. A further work also showed a better prediction of histologic activity (84.5% vs. 60%) and disease extent (89% vs. 52%) by DCE compared to white‐light endoscopy (WLE). ¹² The evidence is still scarce, and more studies are needed to confirm these preliminary findings.

Dye‐less virtual electronic chromoendoscopy

Virtual electronic chromoendoscopy can accurately identify mucosal inflammation and strongly correlates with histology. Iacucci et al. ¹³ found that iSCAN‐OE can still find mucosal and vascular abnormalities in 31% and 41% of Mayo 0 patients, respectively. Also, they described a strong correlation between endoscopic assessment of mucosal inflammation in UC with iSCAN‐OE and histology (r 0.7 and 0.61 by Extension, Chronicity, Activity Plus [ECAP] and Robarts histopathology index [RHI], respectively). These results were confirmed by a further prospective study from the same group, which showed an accuracy of VCE in predicting histologic healing of 95.9%. ¹⁴ Other authors had similar results, indicating an overall agreement between the endoscopic prediction of disease activity and histologic findings of 54% and 90% for high‐definition‐WLE and VCE, respectively. ¹⁵ As concerns NBI, Kudo et al. ¹⁶ described a precise assessment of inflammation in 30 patients with quiescent UC. By evaluating the mucosal‐vascular pattern (MVP), VCE showed a better ability than conventional endoscopy in identifying areas corresponding to increased histologic activity. Areas with obscure MVP revealed significantly increased inflammatory infiltrate and goblet cell depletion at histology compared to distorted and clear MVP (26% vs. 0% and 32% vs. 5%, respectively). Similarly, Danese et al. ¹⁷ showed that areas appearing normal at WLE could demonstrate subtle abnormalities at NBI with consensual inflammatory changes at histology. LCI was superior to WLE in distinguishing between histologic mucosal activity and inactivity. ¹⁸ Also, LCI demonstrated a strong correlation (up to 94% for LCI‐A) with histologic inflammation, as assessed by Matts grading. ¹⁹

The endoscopic scores currently used in clinical practice to grade disease activity in UC, i.e. Mayo endoscopic subscore (MES) and Ulcerative Colitis Endoscopic Index of Severity (UCEIS), were developed using WLE. However, mucosal and vascular features visible only with VCE led to new endoscopic scores. Notably, the Paddington International Virtual Chromoendoscopy (PICaSSO) score, represented in Figure 1, was initially validated and developed using the iSCAN platform, showing a good accuracy in assessing histologic abnormalities: 72% by RHI and 83% by ECAP. ²⁰ This score was then reproduced on NBI and BLI/LCI platforms, confirming a strong correlation with histology. ²¹ NBI‐assessed PICaSSO had a correlation of 0.83 and 0.79 with RHI and the Nancy histological index (NHI), respectively, and an accuracy for HR of 0.80 by RHI and 0.82 by NHI. Similarly, a correlation of 0.63 with RHI and 0.65 with NHI was found for LCI/BLI, with an accuracy for HR of 0.83 by RHI and 0.79 by NHI.

Figure 1.

Paddington International Virtual Chromoendoscopy (PICaSSO) score. This figure schematically represents the novel score based on virtual chromoendoscopy called PICaSSO. The PICaSSO score grades architectural changes of the mucosa (including microerosions, cryptal abscess, and ulcers) and of vessels (such as dilatated or crowded vessels and bleeding). It ranges from 0 to 15 and endoscopic remission is defined by a score equal to or <3. Created with Biorender.com.

Probe‐based confocal laser endomicroscopy

Probe‐based confocal laser endomicroscopy enables an in‐vivo dynamic structural and functional assessment of the intestinal barrier, going beyond the classical concept of endoscopic remission and HR. Distorted/elongated crypts, epithelial gaps, shedding of cells, and fluorescein leakage are the main abnormalities in IBD patients evaluated with pCLE. These features have been demonstrated to correlate adequately with a histologic assessment of inflammation. Indeed, pCLE can identify residual inflammatory changes even in quiescent IBD patients. ²² Also, Li et al. ²³ showed that more than 50% of 73 patients with UC and normal‐appearing mucosa at WLE had acute inflammation on histology. In contrast, all patients in remission according to pCLE had consensual HR. Notably, the recent Endoscopic Remission, Histologic Remission and Barrier Healing for Predicting Disease Behavior in IBD (ERIca) trial assessed the barrier function by pCLE in 181 IBD patients in clinical remission. ²⁴ A combined endoscopic and histologic healing by RHI was present in 44.4% of UC patients and 49% of CD patients, while barrier healing was found only in around 25% of patients.

Different scores have been created to define disease activity through pCLE. The Watson score, developed to evaluate the degree of local barrier dysfunction in IBD according to the amount of cell shedding and the intensity of luminal fluorescein signal, demonstrated a good correlation with histology. ^{25 , 26} Neumann et al. ²⁷ proposed another pCLE score for CD, based on crypts and goblet cells number, which was able to predict HR in apparently noninflamed areas with an accuracy of 87%. A further score, assessing crypt number, crypt lumen deformity, crypt lumen leakage, and vascular leakage, showed high accuracy (94.4%) in defining mucosal changes before and after biological therapy in UC and demonstrated a good correlation with histologic and endoscopic scores. ²⁸

Endocytoscope

Different endocytoscope scores based on crypt architecture, microvessels, and inflammatory cell infiltration have been developed, and the correlation between disease activity assessed through endocytoscope and histology was evaluated. ²⁹

The score developed by Iacucci et al. ²⁹ showed a strong correlation with endoscopic (UCEIS r 0.74, PICaSSO r 0.67) and histologic scores (RHI r 0.89 and NHI r 0.86). Furthermore, the ErLangen Endocytoscopy in ColiTis (ELECT) score developed by Vitali et al. ³⁰ strongly correlated with histology (RHI r 0.7 and NHI r 0.73) and was better than WLE in grading microscopic disease activity. Nakazato et al. ³¹ found an agreement between endocytoscope‐defined remission and histology equal to 0.72, with an accuracy of their score in defining HR of 0.86.

Of note, the combination of endocytoscope and NBI, studied in UC patients by Maeda et al., ³² showed a better correlation with histology than conventional endoscopy (r 0.871 vs. 0.665) and a significantly higher accuracy for the detection of acute inflammation (92.3%).

In summary, advanced endoscopic tools have shown superior ability compared to conventional endoscopy in assessing inflammation in IBD (Table 1). Notably, VCE stands out as a valuable instrument in clinical practice, demonstrating promising predictive abilities for grading disease activity. The emerging technologies of pCLE and endocytoscopy provide the opportunity to assess both barrier structure and function, thereby challenging the concept of MH.

Table 1.

Overview of the main studies evaluating the ability of advanced endoscopic techniques to assess disease activity in inflammatory bowel disease (IBD)

Author	Study design	Patients	Technique	Comparator	Outcome	Ref
Dye and dye‐less chromoendoscopy
Kiesslich et al.	Prospective	165 UC	DCE	WLE	Accurate diagnosis of extent and degree of inflammation	12
Iacucci et al.	Prospective	41 UC and 9 controls	iSCAN‐OE	WLE	Accurate identification of mucosal inflammation and good correlation with histology	13
Iacucci et al.	Prospective	82 UC	iSCAN	HD‐WLE	Accurate prediction of histologic healing by RHI and ECAP	14
Neumann et al.	Prospective	78 IBD	iSCAN	HD‐WLE	Significant improvement in the diagnosis of severity and extent of mucosal inflammation	15
Kudo et al.	Prospective	30 UC	NBI	WLE	Good determination of the grade of inflammation in quiescent disease	16
Danese et al.	Prospective	14 IBD	NBI	WLE	In‐vivo imaging of intestinal angiogenesis	17
Kanmura et al.	Prospective	21 UC	LCI	WLE	Accurate visualization and evaluation of mucosal inflammation	18
Uchiyama et al.	Prospective	52 UC	LCI	WLE	Strong correlation with histology	19
Iacucci et al.	Retrospective	20 UC videos	iSCAN	–	Development of PICaSSO: high accuracy in assessing histologic abnormalities by RHI and ECAP	20
Cannatelli et al.	Prospective	159 UC	NBI and LCI/BLI	–	PICaSSO reproducibility with all VCE platforms: high accuracy in assessing histologic abnormalities by RHI and NHI	21
CLE
Li et al.	Prospective	73 UC	CLE	WLE	Reliable real‐time assessment of inflammation activity Crypt architecture, microvascular alterations and fluorescein leakage promising markers	23
Rath et al.	Prospective	181 IBD	CLE	WLE	Barrier healing associated with decreased risk of disease progression in clinical remission Superior predictive performance than endoscopic and histologic remission	24
Hundorfean et al.	Prospective	23 UC	CLE	–	Accurate assessment of mucosal healing based on eMHs Prediction of long‐lasting clinical outcome	28
Endocytoscope
Iacucci et al.	Prospective	29 UC	Endocytoscope	–	Endocytoscope scoring system correlated strongly with RHI and NHI	29
Vitali et al.	Prospective	46 UC	Endocytoscope	WLE	Accurate grading of microscopic inflammation Accurate prediction of clinical outcome ELECT score correlated strongly with RHI and NHI	30
Nakazato et al.	Retrospective	64 UC	Endocytoscope	WLE	Accurate assessment of histologic healing	31
Maeda et al.	Retrospective	52 UC	Endocytoscope	Conventional endoscopy	Strong correlation with histologic inflammation	32

BLI, blue light imaging; CLE, confocal laser endomicroscopy; DCE, dye‐based chromoendoscopy; ECAP, Extension, Chronicity, Activity Plus; ELECT, ErLangen Endocytocopy in ColiTis; eMHs, endomicroscopic mucosal healing score; HD, high‐definition; LCI, linked color imaging; NBI, narrow band imaging; NHI, Nancy histological index; OE, optical enhancement; PICaSSO, Paddington International Virtual Chromoendoscopy Score; RHI, Robarts histopathology index; UC, ulcerative colitis; VCE, virtual electronic chromoendoscopy; WLE, white light endoscopy.

Image‐enhanced endoscopy to assess and characterize dysplasia

Inflammatory bowel disease patients are at an increased risk of developing colorectal cancer. ³³ Recent population studies showed adjusted hazard ratios of 1.66 and 1.40 in UC and CD, respectively. ^{34 , 35}

Inflammatory bowel disease dysplasia is often found in flat mucosal abnormalities. Therefore, precision endoscopy is essential to detect these early lesions in the dysplasia‐carcinoma pathway. The SCENIC consensus statement notes that chromoendoscopy is the most sensitive modality for detecting dysplasia in colitis. ³⁶ For this reason, endoscopic surveillance with either DCE or VCE is considered the mainstay of dysplasia detection and characterization in IBD. ³⁷ In addition, novel endoscopic tools, such as pCLE and endocytoscope, can increase subtle dysplasia detection and guide endotherapy, even if they have not yet been adopted widely in routine clinical practice.

Dye and dye‐less chromoendoscopy

Dye‐based chromoendoscopy enhances mucosal irregularity and the border of lesions. Early studies have supported DCE over standard WLE with random biopsies to increase dysplasia detection in the DCE groups. ^{12 , 38 , 39} A recent meta‐analysis revealed that DCE was more effective than WLE standard endoscopy in dysplasia detection. ⁴⁰ Notably, Soetikno et al. ⁴¹ showed a pooled incremental yield of DCE over WLE for the detection of any grade of dysplasia per patient of 7%.

Several studies have demonstrated the value of VCE in comparison to WLE, showing similar intraepithelial neoplasia detection rates with less biopsy collection and statistically significant reduced withdrawal time. ⁴² Although several recent meta‐analyses showed the superiority of DCE and VCE over standard WLE in detecting dysplasia, the comparison of diagnostic yield between DCE and VCE remains a topic of discussion. ^{43 , 44} El‐Dallal et al. found no statistical difference between VCE and DCE (risk ratio 0.72) per dysplasia analysis. ⁴⁴

European and American guidelines recommend the use of either DCE or VCE with targeted biopsies for dysplasia detection and assessment. ^{37 , 45 , 46 , 47}

Probe‐based confocal laser endomicroscopy

A first work by Kiesslich et al. ⁴⁸ demonstrated a 4.75 times higher neoplasia detection rate with pCLE than conventional colonoscopy, with high sensitivity (94.7%) and specificity (98.3%). However, a subsequent multicentric study, terminated early due to critical equipment failure, showed pCLE reduced diagnostic yield, with a sensitivity of 42.9% and specificity of 92.4%. ⁴⁹ Notwithstanding, a recent study clarified the diagnostic utility of pCLE‐aided surveillance in IBD patients with primary sclerosing cholangitis, showing a high sensitivity (89%), specificity (96%), and accuracy (96%). ⁵⁰ More recently, Ohmiya et al. ⁵¹ evaluated the ability of pCLE to differentiate among UC‐associated neoplasia, sporadic adenoma, and circumscribed regenerative lesions. A high accuracy (92%) of the combination of specified pCLE features, namely, back‐to‐back orientation of crypts and dark trabecular architecture, was described for carcinoma or dysplasia detection. These data suggested that pCLE should be helpful for a detailed characterization of lesions after the detection made with chromoendoscopy.

Endocytoscope

To date, few case reports have suggested the potential role of an endocytoscope in IBD dysplasia assessment. ⁵² Also, a recent pilot study by Kudo et al. showed that the combination of pit pattern with endocytoscope‐based evaluation of irregularly‐formed nuclei (so‐called EC‐IN‐PIT strategy) had a better accuracy than pit pattern alone (88% vs. 67%) to predict UC‐associated neoplasia. ⁵³ Further studies are needed to confirm this particular applicability in clinical practice. Nonetheless, some data can be derived from studies on sporadic lesions: a recent prospective study on 75 bowel lesions showed an excellent ability of the endocytoscope in distinguishing neoplastic vs. nonneoplastic colorectal lesions and adenoma vs. invasive adenocarcinoma, with good correspondence to traditional microscopy (overall accuracy 93.3%). ⁵⁴

Advanced techniques, especially chromoendoscopy, enhance the detection and characterization of dysplastic lesions in IBD (Table 2). By providing more detailed lesion characterization and improving tissue sampling, these tools are helpful for the early detection and management of colitis‐related neoplasia (Fig. 2). The main challenge persists in diagnosing low‐grade dysplasia, which often remains modest on both conventional and magnified endoscopy, and also poses difficulties for pathologists.

Table 2.

Overview of the main studies evaluating the ability of advanced endoscopic techniques to assess and characterize dysplasia

Author	Study design	Patients	Technique	Comparator	Outcome	Ref
Dye and dye‐less chromoendoscopy
González‐Bernardo et al.	Prospective	129 IBD	DCE	iSCAN1	No differences in detection rate of neoplastic lesions	9
Alexandersson et al.	Prospective	305 IBD	HD‐DCE	HD‐WLE	Better dysplasia detection	10
Kiesslich et al.	Prospective	165 UC	DCE	WLE	Early detection of intraepithelial neoplasia	12
Marion et al.	Prospective	102 IBD	DCE	WLE	Improved dysplasia yield	39
Feuerstein et al.	Meta‐analysis	1562 IBD	DCE	HD‐WLE and SD‐WLE	RCTs: small benefit over SD‐WLE, but not over HD‐WLE for dysplasia detection Non‐RCTs: benefit over SD‐WLE and HD‐WLE for dysplasia detection	40
Soetikno et al.	Meta‐analysis	665 IBD	DCE	WLE	Increase in detection of any dysplasia	41
Kandiah et al.	Prospective	188 IBD	iSCAN‐OE	HD‐WLE	No differences in detection rate of neoplastic lesions	42
Iannone et al.	Meta‐analysis	2638 IBD	iSCAN and SD‐WLE	DCE	Low yields for neoplasia identification	43
CLE
Kiesslich et al.	Prospective	161 UC	CLE	Conventional colonoscopy	Increased neoplasia diagnostic yield Reduced need for biopsy	48
Wanders et al.	Prospective	61 CD	CLE	DCE	Good accuracy in differentiating neoplastic from nonneoplastic lesions	49
Dlugosz et al.	Prospective	69 IBD	CLE	HD‐WLE	High accuracy for dysplasia detection Good performance for differentiating neoplastic from non‐neoplastic mucosa	50
Ohmiya et al.	Prospective	12 UC	CLE	NBI and DCE	High accuracy for carcinoma or dysplasia based on back‐to‐back crypts orientation and dark trabecular architecture	51
Endocytoscope
Kudo et al.	Retrospective	62 UC	Endocytoscope	WLE	Good diagnostic ability in predicting UC‐associated neoplasia	53

CD, Crohn's disease; CLE, confocal laser endomicroscopy; DCE, dye‐based chromoendoscopy; HD, high‐definition; IBD, inflammatory bowel disease; NBI, narrow band imaging; OE, optical enhancement; RCT, randomized controlled trial; SD, standard definition; UC, ulcerative colitis; WLE, white light endoscopy.

Figure 2.

Advanced endoscopy for detection and assessment of inflammatory bowel disease (IBD)‐related colonic lesions. This figure highlights the role of advanced endoscopy in the detection and assessment of IBD‐related colonic lesions. Images provided were obtained through dye‐based chromoendoscopy (DCE), virtual electronic chromoendoscopy (VCE), confocal laser endomicroscopy, and endocytoscopy. pCLE, probe‐based confocal laser endomicroscopy. Created with Biorender.com.

ENDOSCOPY‐ENABLED PREDICTION OF CLINICAL OUTCOME AND RESPONSE TO THERAPY

Advanced endoscopic techniques offer considerable value in achieving a “pinpoint” diagnosis of inflammation. VCE, when combined with pCLE or the endocytoscope, enhances the characterization of the intestinal mucosa, vascular details, and the ultrastructure of the intestinal barrier. This comprehensive approach enables simultaneous and comprehensive assessment of the most representative areas of the intestinal mucosa, leading to improved outcome prediction. Furthermore, the application of endoscopy in analyzing molecular signaling pathways (so‐called molecular endoscopy) would allow for predicting responses to advanced therapies. ⁵⁵

Endoscopic prediction of outcome through the eyes of enhanced images

Concerning VCE and outcome prediction, data available in the literature are related only to UC. The relationship between magnified NBI endoscopy and prognosis was first evaluated in 52 UC patients in clinical remission with an MES of 0 or 1. ⁵⁶ These patients were assessed with NBI at enrollment and after a 1‐year follow‐up, the findings were stratified according to vessel appearance. Differences in NBI findings reflected different prognoses: patients with vessels shaped like a bare branches pattern showed a higher risk of relapse (odds ratio 14.2) than those showing a honeycomb‐like blood vessels pattern. Also, the VCE mucosal and vascular assessment through the PICaSSO score has shown a good prediction of outcomes. A large prospective real‐life international study evaluating a cohort of 307 UC patients showed that a PICaSSO score ≤3 predicted major adverse outcomes (MAOs) at 6 and 12 months better than a PICaSSO >3 (hazard ratio 0.19 and 0.22, respectively) and similarly to HR. ⁵⁷ In the external validation study, a PICaSSO score ≤3 was associated with increased relapse‐free survival rates than a score >3 (hazard ratio 0.189). ⁵⁸ Moreover, the endoscopic remission by VCE‐PICaSSO alone has been similar to combined endoscopic and HR for predicting specified clinical outcomes at 12 months in UC patients. ⁵⁹ Notably, a recent work by Hayashi et al. ⁶⁰ evaluated the usefulness of texture and color enhancement imaging (TXI) (Olympus) in predicting UC relapse. A TXI score of 2 resulted in a risk factor for disease relapse (hazard ratio 4.16).

As concerns the pCLE, several studies have shown its ability to predict outcomes in IBD. In previous studies focusing only on CD patients, pCLE features such as fluorescein leakage, microerosion, focal cryptitis, and crypt architectural abnormality have been the early predictors of relevant clinical outcomes at 1‐year follow‐up. ^{61 , 62} Similarly, past studies on UC patients highlighted a correlation between crypt architecture/fluorescein leakage and disease flare‐ups. ^{63 , 64} In the previously mentioned ERIca trial, barrier healing showed superiority compared to endoscopic and HR for predicting MAO‐free survival. ²⁴

The endocytoscopic ability to predict clinical outcomes has been assessed in UC; however, no data are available for CD. A first pilot study in 2015 demonstrated a correlation between the relapse rate and the endocytoscopy system score (ECSS) in 26 UC patients in remission, and only patients with a higher ECSS relapsed. ⁶⁵ A similar study was conducted in 32 patients with mild–moderate UC stratified according to endocytoscope findings. ⁶⁶ A normal endocytoscopy appearance, defined as a regular arrangement of round to oval pits, was associated with no relapse. In contrast, pit alterations were related to relapse episodes at 60 month follow‐up (P < 0.005). Subsequent studies involving a wider number of UC patients have confirmed the ability of some endocytoscopic features, such as intramucosal capillary network changes, crypt architecture abnormalities, and goblet appearance, to predict relapse. ^{67 , 68} Finally, the endocytoscopic remission defined as an ELECT score ≤2 showed an accuracy almost comparable to histopathology (67.7%) in forecasting the occurrence of clinically relevant MAOs. ³⁰

Endoscopic prediction of response to therapy through the eyes of enhanced images

At present, only about 40–60% of patients treated with biologic therapy respond to treatment, and a prognostic indicator of response to treatment in IBD is still missing. ^{69 , 70} The possible application of molecular endoscopy in predicting a response to biologics has been tested. In particular, the in vivo topical administration of fluorescent antibody (anti‐tumor necrosis factor [TNF]) in 25 CD patients led to the detection of intestinal membrane‐bound (m)TNF⁺ immune cells during pCLE. ⁷¹ Patients with a high number of mTNF⁺ cells showed a higher response rate (92%) at week 12 and a sustained response over 1‐year follow‐up. A subsequent pilot study assessed in five CD patients the correlation between the number of α4β7‐expressing cells in the mucosa, assessed through ex vivo pCLE, and vedolizumab therapeutic efficacy. ⁷² Two patients with detectable α4β7‐expressing cells responded to therapy, while no α4β7‐expressing cells were observed in the other three patients who were nonresponders. These findings pave the way for the potential use of molecular endoscopy for personalized medicine. However, larger cohort studies are needed to validate this technique.

To summarize, advanced endoscopic techniques have shown a remarkable ability to predict clinical outcomes in IBD. Furthermore, recent studies have provided promising evidence for the potential prediction of response to biologics through molecular endoscopy, paving the way for personalized and targeted approaches in IBD management.

AI‐ENABLED ENDOSCOPY PRECISION IN DIAGNOSIS AND PREDICTION

AI‐enabled assessment of inflammation and prediction of outcome

Several groups have developed deep‐learning models to grade endoscopic disease activity objectively using MES, ^{73 , 74 , 75} UCEIS, ^{76 , 77} or both scores ^{78 , 79} and showed excellent diagnostic performance and strong concordance with experts.

Moreover, some of the models were trained and tested against expert reading based on high‐quality images and videos from prospective multicenter clinical trials, showing the potential of AI to replace central expert readers in future clinical trials. ^{75 , 78}

An operator‐independent computer‐based tool called red density (RD) was developed by Bossuyt et al. ⁸⁰ The RD is based on integrating pixel color data from the redness color map and vascular pattern detection. A moderate correlation between the RD score and endoscopic and histologic activity in UC was found.

Takenaka et al. ⁷⁶ developed a deep neural network for UC (DNUC) to differentiate endoscopic remission (UCEIS = 0) from HR (Geboes score ≤3). When tested prospectively on endoscopic videos from 770 patients, DNUC showed high accuracy, with a sensitivity of 97.9% and specificity of 94.6%, in predicting HR and strongly correlated with experts scoring of UCEIS. ⁷⁴ DNUC also predicted clinical outcomes at 12 months. ⁸¹

The first AI system based on the PICaSSO score was developed by Iacucci et al. ⁸² to distinguish endoscopic remission/activity, and predict histology and clinical outcomes. The model showed an area under the receiver operating characteristic curve (AUROC) of 0.94 in predicting endoscopic remission and an 80–85% diagnostic accuracy in predicting HR. In addition, the AI‐PICaSSO model was found to predict clinical outcomes at 12 months.

Further, a computer‐aided diagnosis (CAD) system based on endocytoscopy was introduced by Maeda et al. ⁸³ to predict endoscopic and HR. The CAD system showed a diagnostic accuracy of 91%. The authors further developed the CAD system (EndoBRAIN‐UC; Cybernet Systems, Tokyo, Japan) using images from endocytoscope and ultramagnified NBI findings to classify patients into either AI‐healing or AI‐histologically active groups. ⁸⁴ The model predicted a clinical relapse rate at 12 months (28.4% in AI‐histologically active group vs. 4.9% in the AI‐healing group, P < 0 0.001).

As concerns CD, few studies evaluated the possible application of AI on imaging, mainly focusing on capsule endoscopy and showing a good ability to diagnose small bowel ulcers and bleeding. ⁸⁵

AI‐enabled prediction of therapeutic response

The putative role of AI‐aided endoscopy in the prediction of response to therapy was recently evaluated in the Endo‐Omics study: computer‐aided imaging analysis was developed using in vivo pCLE images and ex vivo molecular labeling of mucosal biopsies to predict response to biologics. ⁵⁵ In vivo, vessel tortuosity, crypt morphology, and fluorescein leakage findings predicted response in UC (AUROC 0.93, accuracy 85%) and CD (AUROC 0.79, accuracy 80%). Also, ex vivo increased binding of a labeled biologic predicted response in UC (AUROC 83%, accuracy 77%). Finally, genes predictive of response to therapy were determined, and a panel of genes, including those related to chemotactic pathways, showed a good prediction of therapeutic response. That study requires further validation and replication in larger studies.

AI‐enabled prediction of IBD‐associated dysplasia

There are only two case reports on detecting IBD‐associated dysplasia through an AI system (EndoBRAIN‐EYE; Cybernet Systems). ^{86 , 87} Also, computer‐aided systems able to detect adenoma and predict histology of colorectal polyps have been developed, but only in non‐IBD populations. ⁸⁸

Yamamoto et al. ⁸⁹ tested an AI system for characterizing neoplasia occurring in IBD. The model classified lesions into two groups: “adenocarcinoma/high‐grade dysplasia” and “low‐grade dysplasia/sporadic adenoma/normal mucosa.” When compared against experts and nonexperts, the AI diagnostic accuracy of the AI model was higher (nonexperts, 77.8%; experts, 75.8%; AI model, 79.0%). However, this is a pilot model and external validation is still required.

In conclusion, the integration of AI in advanced endoscopy holds great promise in IBD. ⁹⁰ It is expected to revolutionize the assessment of inflammation and improve diagnostic accuracy. AI will detect subtle changes that may not be visible to the human eye and reduce interobserver variability. Additionally, it will play a role in real‐time decision‐making during endoscopic procedures, leading to increased lesion detection rates and providing predictive insights into clinical outcomes. The AI‐driven analysis will guide therapeutic strategies, enabling the practice of precision medicine.

ENDOSCOPY PRECISION GUIDING ENDOTHERAPY

The accuracy in endoscopic detection, assessment, and resection of inflammation‐related lesions in IBD has been of increasing interest in the last few years, aiming at a surgery‐sparing approach.

Extensive efforts have been made to develop new classifications of IBD‐related dysplastic lesions suitable for WLE, VCE, and DCE ^{36 , 91} and able to guide endotherapy. Notably, the Kudo pit pattern and the Frankfurt Advanced Chromoendoscopic IBD LEsions classification, assessing lesion morphology, vessel architecture, inflammation, and surface pattern, have shown a good prediction of dysplasia. ^{91 , 92 , 93} Also, the European Crohn's and Colitis Organisation recently suggested the implementation of a standardized reporting system incorporating lesion features to optimize the characterization of lesions and guide subsequent management. ⁹⁴ The report should include the assessment of five key aspects, referred to as the five ‘S’: shape, size, site, surface, and surrounding area. However, a wide applicability in daily clinical practice still needs to be established.

Endoscopic mucosal resection and endoscopic submucosal dissection are commonly used for the resection of IBD‐associated lesions. ⁹⁵ Endoscopic assessment and clear margin definition are crucial to guide these therapeutic approaches, since endoscopically unresectable dysplastic lesions may be referred for colectomy. ³⁷ In addition, a more precise assessment of lesion and margins could better guide radical resection and reduce or prevent the risk of recurrence. An image‐enhanced endoscopy could better assess several features that define a lesion as unresectable, such as nondefined margins, submucosal invasion, and flat neoplastic change with distorted pit patterns adjacent to the lesion. ^{41 , 96 , 97}

Further studies are needed to assess the potential benefits of advanced endoscopic tools and AI‐assisted techniques in assessing and resecting IBD‐related lesions, driving towards their use in clinical practice.

CONCLUSION

In conclusion, the remarkable potential of new advanced endoscopic techniques highlights their importance in clinical practice, aiming to enhance the definition of endoscopic, mucosal, and barrier healing. The improved detection and assessment of dysplasia are also promising for guiding endotherapy and managing IBD‐related lesions. The forthcoming integration with AI presents an opportunity for standardizing the evaluation obtained through these techniques, enabling better prediction of disease progression. The combination of novel endoscopic tools with pioneering AI systems and molecular endoscopy will transform precision medicine in IBD and IBD‐associated dysplasia from a futuristic goal into a tangible reality.

CONFLICT OF INTEREST

Author M.I. is an Associate Editor of Digestive Endoscopy. The other authors declare no conflict of interest for this article.

FUNDING INFORMATION

None.