Can Spectral CT Complement Biopsy for MSI/dMMR Assessment in Esophagogastric Junction Adenocarcinoma?

See also the article by Wu and Lin et al in this issue.

Ramiro Méndez, MD, is chief of the abdominal radiology section at the Hospital Clínico San Carlos and associate professor of radiology at the Universidad Complutense de Madrid. He is a consultant of abdominal imaging at Hospital Nuestra Señora del Rosario. His clinical and research areas of interest are focused on abdominal and pelvic MRI and CT. He has been president of the Spanish Society for Abdominal Imaging and is on the editorial board of European Radiology.

Ángel Nava, MD, is an abdominal radiologist at the Hospital Clínico San Carlos and Hospital Nuestra Señora del Rosario, in Madrid, Spain. His clinical and research interests focus on abdominal and pelvic MRI and CT, with a growing interest in artificial intelligence and three-dimensional printing applications in radiology. He also teaches medical image processing to telecommunications engineering students.

Esophagogastric junction adenocarcinoma (EGJA) is a distinct clinical entity, different from both gastric and esophageal cancers. Its prognosis is particularly poor, with about half of patients with nonmetastatic disease experiencing recurrence after surgery despite multimodal treatment. Over the last decade, molecular profiling has assumed increasing clinical importance in gastroesophageal cancers, reflected in newer molecular classification schemes for this tumor type (1). Among the available biomarkers, few have as direct an impact on prognosis and treatment selection as microsatellite instability (MSI) and deficient mismatch repair (dMMR) status. Patients with MSI/dMMR tumors have a better prognosis and achieve high response rates to immune checkpoint inhibitors, whereas patients with microsatellite-stable (MSS)/proficient mismatch repair (pMMR) tumors generally do not respond to immunotherapy and instead receive conventional chemotherapy. Objective response rates to immunotherapy in MSI/dMMR tumors are high, with up to 50% of patients achieving a pathologic complete response in the surgical specimen, and some groups have even proposed organ-preserving strategies for selected responders (2).

Due to these important implications for clinical management, routine testing for MSI/dMMR is recommended by most clinical practice guidelines in patients with EGJA (3,4). Approximately 15% (range, 8%–22%) of patients with nonmetastatic disease have MSI/dMMR tumors, a substantially higher prevalence than in patients with metastatic disease (3%–5%) or those with purely esophageal adenocarcinoma (5).

Variability in the reported prevalence of MSI/dMMR in nonmetastatic EGJA among different published series may reflect geographical differences, but it could also be related to variability in the testing methods or reveal limitations of endoscopic sampling in accurately representing the characteristics of the entire tumor. Endoscopic biopsy samples only the superficial mucosa of a tumor that may be spatially heterogeneous. The clinical significance of this heterogeneity remains unclear. Several series have reported high concordance between biopsy and surgical specimen MSI/dMMR classification, but at least one dedicated multicenter study in resectable EGJA found discordant results in nearly one in five paired cases (6). Multiregional sampling studies have also described areas of MSS tissue within otherwise MSI-high/dMMR tumors, pointing to genuine, although not universal, intratumoral heterogeneity (7). Although immunohistochemistry or polymerase chain reaction/next-generation sequencing analysis of endoscopic biopsy specimens remains the standard for selecting initial therapy, a noninvasive method for assessing MSI/dMMR status could play a complementary role.

In this issue of Radiology: Imaging Cancer, Wu et al (8) present a nomogram based on simple clinical and quantitative spectral (multienergy) CT parameters—sex, clinical N stage, CT attenuation on 40-keV virtual monoenergetic images, and normalized iodine density in the venous phase—to predict MSI/dMMR status in EGJA. Spectral CT parameters have already been linked to biomarkers such as MSI status and TP53 expression in gastric cancer (9), but this study stands out primarily because of its scale and validation design. A total of 511 patients with EGJA from two centers were divided into training, validation, external test, prospective test, and neoadjuvant chemotherapy cohorts, with spectral CT features correlated with MSI/dMMR status on the surgical specimen. The nomogram achieved areas under the receiver operating characteristic curve ranging from 0.86 to 0.91 across all five cohorts. The authors do not propose the model as a replacement for MSI/dMMR testing on the initial biopsy but rather as a complementary safeguard. This is supported by confusion matrix analyses showing that the nomogram correctly reclassified a meaningful proportion of patients misclassified by biopsy in every cohort, including all misclassified patients in the external test set. A combined biopsy-nomogram model was also constructed and demonstrated better performance than endoscopic biopsy alone.

The spectral CT–based nomogram also stratified disease-free survival, adding a prognostic dimension beyond biomarker prediction alone. Moreover, the nomogram was evaluated for predicting tumor response in the neoadjuvant cohort and outperformed endoscopic biopsy.

The physical rationale underlying these spectral CT parameters is sound and, importantly, mechanistically transparent rather than a “black box.” Iodine density and attenuation on low-energy virtual monoenergetic images reflect contrast media uptake, which is related to tumor vascularity, permeability, and interstitial space (10). The finding that MSS/pMMR tumors show higher iodine density than MSI/dMMR tumors is consistent with previously described differences in angiogenesis and stromal composition between these two subtypes. This interpretability is an important strength compared with deep learning– or radiomics-based classifiers, which may achieve similar performance but provide little insight into how predictions are generated—a substantial barrier to clinical trust and regulatory acceptance.

The authors went one step further by simulating conventional 120-kVp images from 70-keV monoenergetic reconstructions. The spectral parameters outperformed the conventional CT approximation, confirming the added value of spectral acquisition while also suggesting that conventional CT attenuation measurements may retain some, albeit reduced, predictive value for centers without spectral CT scanners.

Several limitations should temper how quickly this approach moves into routine clinical practice. Despite the study’s retrospective-prospective, two-center design, all examinations were performed using the same spectral CT scanner, limiting generalizability to other CT vendors. The reported prevalence of MSI/dMMR is higher than in other published series. This likely reflects geographical and referral biases, which may have inflated diagnostic performance. Tumor measurements were obtained from a single representative axial section rather than volumetric assessment of the entire tumor; the authors acknowledge that this introduces an element of manual selection, although interobserver agreement was good. The number of patients actually reclassified by the nomogram in each cohort was modest, and follow-up for the disease-free survival analysis was relatively short. Therefore, the prognostic findings should be considered hypothesis-generating rather than definitive. Finally, because MSI/dMMR status is itself prognostic, it is genuinely difficult to separate the nomogram’s independent prognostic value from that of the molecular subtype it predicts—a limitation the authors openly acknowledge.

None of these limitations diminish the potential clinical impact of the study. CT is already performed in nearly every patient with EGJA for staging purposes, so a model based on parameters readily obtained from that same examination has an appealing “no additional cost” rationale. This contrasts with MRI-based approaches, which are rarely part of routine gastroesophageal cancer staging. Adoption of the CT-based nomogram in clinical practice could be facilitated by a simple spreadsheet or online calculator. If discrepancies between biopsy findings and CT-based predictions prompt pathologists or oncologists to re-evaluate a case, as illustrated by the discordant patients presented by the authors, the value of this tool may lie less in replacing histopathology and more in identifying cases that warrant a second look before an irreversible treatment decision is made.

Before this vision can be realized, several next steps are needed: external testing across different spectral CT technologies and vendors; larger prospective cohorts, ideally including more patients with EGJA undergoing neoadjuvant therapy, with longer follow-up to confirm the prognostic signal; and, ideally, a study directly comparing biopsy findings, the CT-based nomogram, and combined decision-making against hard treatment-outcome end points. If these steps confirm the present findings, spectral CT–derived iodine and attenuation metrics may become a practical complement—not a replacement—to the preoperative molecular characterization of EGJA, adding a noninvasive layer of confidence to a decision that currently depends on a very small piece of tissue.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials.