TO THE EDITOR
We read with great interest the article by Ding et al[1], which was published in the World Journal of Gastroenterology. That study reported a noninvasive radiomic model, establishing a novel imaging biomarker for the evaluation of mucosal healing (MH) in patients with Crohn’s disease (CD). This retrospective investigation employed radiomic features from computed tomography enterography (CTE) to build a prediction model. Concurrently, patient demographics were analyzed to create a clinical model. A integrated predictive model was subsequently developed by combining radiomic signatures with clinical parameters. This composite tool exhibited high diagnostic accuracy for identifying MH in CD patients, with area under the receiver operating characteristic curve values reaching 0.961 in the training set and 0.958 in the validation cohort. The study suggests that this radiomic-clinical model could serve as a reliable imaging biomarker and a viable non-invasive method for assessing MH, potentially reducing the reliance on endoscopic evaluation in CD management. In light of recent literature, in this letter, the role of cross-sectional imaging in disease monitoring in current clinical practice is discussed, and future research directions that will define its role in the management of CD are explored.
The treat-to-target strategy optimizes long-term management of CD
CD is a chronic inflammatory disorder of the gastrointestinal tract marked by progressive transmural inflammation, which leads to destructive full-thickness lesions[2]. Dynamic monitoring of disease activity and personalized treatment are essential for improving long-term prognosis. The International Organization for the Study of Inflammatory Bowel Diseases advocates a “treat-to-target” strategy for clinical management, stratifying treatment targets into short-term, intermediate, and long-term goals, defined as clinical response, clinical remission, and MH, respectively[3]. Since the clinical response or remission is insufficient to alter the natural course of the disease, MH, evaluated via endoscopy, has been established as a long-term treatment target[4,5]. Although not currently a formal target, transmural healing (TH), assessed by cross-sectional imaging, should be utilized as an adjunct to endoscopic MH to represent a deeper level of healing in patients with CD[3].
MH predicts better long-term results in CD
Achieving MH is widely recognized as being associated with improved long-term results in patients with CD. Attaining this therapeutic target is correlated with reductions in steroid dependency, hospitalizations, and surgical interventions and may also mitigate the risk of cumulative bowel damage[2]. Critically, the persistence of mucosal inflammation, even during clinical remission, is linked to an increased risk of disease-related complications, disease flares, and subsequent surgeries. Endoscopy serves as the gold standard for assessing MH, as it enables direct evaluation of the intestinal mucosal inflammatory status. Currently, several endoscopic scoring systems have been developed and validated for detecting MH, including the simple endoscopic score for CD and the CD endoscopic index of severity. According to a systematic review, the achievement of MH is endoscopically defined by a simple endoscopic score-CD ≤ 2 points or CD endoscopic index of severity < 3 points, with the absence of ulcerations[3]. However, there are several limitations of endoscopy in the evaluation of MH. For example, ileocolonoscopy cannot be used to evaluate extensive small bowel lesions or transmural changes. As an invasive technique, it is also poorly tolerated by patients, particularly those in pediatric and pregnant populations. Additionally, endoscopy has inherent risks, such as intestinal perforation, hemorrhage, and capsule retention, which limits its application in high-risk scenarios. Conversely, cross-sectional imaging techniques enable comprehensive evaluation of transmural and extramural disease throughout the entire bowel, even in patients with strictures or severe inflammation. Several cross-sectional imaging activity scores, such as the magnetic resonance index of activity, Clermont, or bowel ultrasound (BUS) score, have been developed and validated for assessing TH and are strongly correlated with endoscopic measures of mucosal inflammation in the colon and terminal ileum in patients with CD[6,7].
Beyond MH, TH represents a deeper level of healing in CD
CD is a transmural inflammatory bowel disease, and achieving MH alone cannot represent complete resolution of inflammation across the entire bowel wall. Clinically, even if MH is achieved, a subset of patients have persistent transmural inflammation, which remains a risk factor for disease progression. Therefore, TH has emerged in recent years as a potential treatment endpoint in CD management[8-10]. Cross-sectional imaging techniques, including BUS, CTE, and magnetic resonance enterography (MRE), are validated modalities for assessing transmural and extramural disease in patients with CD. A systematic review proposed distinct definitions of TH according to the imaging modality utilized. For MRE or CTE, this status is defined by a bowel wall thickness (BWT) ≤ 3 mm, without contrast enhancement or complications (e.g., abscesses, strictures, or fistulae). Evaluations for this endpoint are typically scheduled at a median of 26 weeks after starting treatment. Using BUS, TH can be assessed as early as 12 weeks after treatment initiation, defined by a BWT ≤ 3 mm and normalized vascularization on doppler imaging[11]. Current evidence indicates no significant differences in diagnostic accuracy among MRE, CTE, and BUS for evaluating TH in patients with CD[12]. Multiple studies have indicated that TH is a better predictor of long-term outcomes than MH is. For example, a prospective study utilizing BUS revealed that TH was correlated with corticosteroid-free clinical remission, delayed drug escalation, and reduced hospitalization risk, whereas MH only delayed drug escalation[13]. Further evidence from BUS, CTE, or MRE assessments confirmed that TH significantly increased steroid-free clinical remission rates while lowering hospitalization and surgical intervention rates compared with MH[9,14,15]. Although TH is established as an independent predictor of long-term outcomes and a potential therapeutic target for CD, several limitations impede its full clinical translation. First, substantial heterogeneity in TH definitions across studies creates ambiguity in clinical decision-making. Simplified TH criteria (e.g., BWT ≤ 3 mm alone) yield higher TH rates but risk underestimating persistent transmural inflammation, potentially leading to premature step-down therapy and increased relapse risk. In contrast, extended criteria (e.g., BWT ≤ 3 mm without hypervascularization signs) reduce observed TH rates while enhancing therapeutic precision and curbing bowel damage progression. Second, the optimal timing for TH assessment lacks consensus, with proposed intervals ranging from 9 weeks to 2 years post-treatment initiation. Given the nonlinear evolution of drug response rates, characterized by suboptimal early-phase response and potential late-phase loss of response, premature assessment risks the underestimation of therapeutic efficacy, while excessively delayed assessment may hinder necessary therapeutic adjustments. Third, conventional imaging evaluations largely rely on radiologists’ subjective assessments, introducing significant interpretative variability. Finally, numerous intrinsic image details remain undetectable through visual inspection alone.
Radiomics serves as a valuable tool for the comprehensive management of CD
Recently, radiomics has emerged as a novel computational methodology for quantifying multi-dimensional features in medical images, enabling the detection of nuanced changes imperceptible to visual examination. With this technique, high-throughput features derived from conventional medical imaging are used to identify non-invasive biomarkers linked to intestinal inflammation, fibrosis, therapeutic response, and prognosis, facilitating clinical decision-making in CD management. Ding et al[1] focused solely on MH, without evaluating TH, and time-consuming and labor-intensive manual segmentation are prone to human error. To address these shortcomings, the same group developed a deep learning radiomics (DLR) model utilizing baseline CTE to noninvasively predict stratified healing encompassing both MH and TH in patients with CD following infliximab therapy. By extracting radiomic and deep learning features from both active lesion walls and mesenteric adipose tissue, the DLR model achieved area under the curves of 0.948 (training), 0.889 (testing), and 0.938 (external validation) for MH prediction, with a diagnostic performance of 0.856 for TH prediction[16]. These results suggest that the model has potential for comprehensive assessment of intestinal healing. In another study, an advanced deep learning model was developed for automated lesion segmentation in CTE images. A machine learning model was subsequently established to extract radiomic features from these segmentation results, facilitating rapid and accurate assessment of disease activity severity in patients with CD[17]. Furthermore, Chirra et al[18] developed and validated a radiomics-based machine learning model using MRE. This model was used to accurately differentiate intestinal strictures with histopathologically severe inflammation from those with milder inflammation, as well as severe fibrosis from less severe fibrosis in CD. Such noninvasive surrogates for pathologic subtypes in stricturing CD could enable more targeted treatment paradigms and patient selection for upcoming clinical trials of antifibrotic therapy. Additional studies have demonstrated that fusion models integrating radiomics with clinical data and radiological scoring systems can be used to predict treatment response to biologics[16,19] and surgical risk in patients with CD[20]. Moreover, using two multidisciplinary machine learning frameworks incorporating clinical features, MRE or CTE signs, endoscopic manifestations, and histopathological results has proven effective in distinguishing CD from intestinal tuberculosis or primary intestinal lymphoma[21,22]. Radiomics facilitates the quantitative analysis of biological heterogeneity extending beyond conventional imaging, with applications including diagnosis, assessment of inflammatory activity and fibrosis, and prediction of treatment response for patients with CD. Nevertheless, current radiomics research in CD predominantly relies on retrospective, single-center cohorts with heterogeneous study designs, limiting the generalizability of the findings. Furthermore, conventional radiomics refers to the extraction of handcrafted features (e.g., texture and morphology) from medical images based on expert-defined algorithms. Its core limitation is the dependence on manual region-of-interest delineation, which is a time-intensive and operator-dependent process. In contrast, DLR leverages convolutional neural networks to autonomously extract hierarchical features directly from raw image data, without predefined handcrafted characteristics or human intervention, thereby reducing observer bias and measurement errors. Such DLR-driven quantitative assessments of bowel morphology demonstrate the potential to standardize measurements, enhance cross-institutional reproducibility, and ultimately improve clinical feasibility in CD management.
Conclusion
In summary, the ultimate therapeutic goal for patients with CD is to achieve sustained deep remission. TH assessed via cross-sectional imaging is more strongly correlated with improved long-term outcomes than endoscopic MH is, positioning it as a superior treatment target. Recent advances in radiomics have enabled the extraction of hidden biological signatures from medical images, offering new pathways for monitoring disease, predicting prognosis, and decision-making. Future research should prioritize the development of deep learning-based automated segmentation tools and establish prospective multicenter data standards to enhance model generalizability. In addition, the integration of radiomics, metabolomics, and genomics is crucial for developing comprehensive predictive models that optimize precision medicine to improve long-term quality of life for CD patients.
