The research, led by Mingjing Lu, Zhongjian Xu, and Tang Hanfei, with senior authors Dehai Lang, Guo Daqiao, and Guofu Wang, took a deliberately dual approach. Rather than betting on a single machine learning paradigm, the investigators built and compared two complementary pipelines. The first was a classical radiomics pipeline, in which hand-engineered quantitative features—describing the shape, texture, and intensity distribution of the plaque—are extracted from manually defined regions of interest on CT angiography images. The second was a deep learning pipeline, in which convolutional neural networks learn discriminative image patterns directly from pixel data without explicit feature engineering. Both pipelines were trained to answer the same clinically critical question: is this plaque histologically stable or vulnerable, as confirmed by pathology after carotid endarterectomy?

The evidence base for the radiomics arm came from a retrospective multi-center cohort of 260 consecutive patients who underwent carotid endarterectomy, meaning that every plaque in the dataset had a gold-standard pathological diagnosis of stability or vulnerability. Of these, 200 patients were used for model development and internal validation, while 60 patients were held out entirely for external validation, a design choice that guards against the optimistic performance estimates that plague many machine learning studies in medicine. For the deep learning arm, the team assembled an independently annotated CT angiography dataset comprising 236 cases and a remarkable 7,394 individual regions of interest, providing the volume of labeled image data needed to train and test convolutional architectures meaningfully.

On the technical side, the radiomics workflow followed rigorous feature-selection discipline. Candidate radiomic features were screened using analysis of variance combined with Kruskal–Wallis testing to identify features that discriminated between stable and vulnerable plaques, followed by correlation filtering to remove redundant measures, and finally least absolute shrinkage and selection operator, or LASSO, regularization to compress the feature set to its most predictive and non-redundant core. Five classifier families were then trained on the selected features: Random Forest, Support Vector Machine, k-Nearest Neighbor, Naïve Bayes, and Logistic Regression. Performance was quantified using receiver operating characteristic analysis and the area under the curve, complemented by calibration analysis, which asks not merely whether a model ranks patients correctly but whether its predicted probabilities match observed reality—a property that matters enormously when model outputs are meant to inform surgical decisions.

The radiomics results were strikingly consistent. On the internal test set, the Random Forest model achieved an AUC of 0.858, followed closely by the Support Vector Machine at 0.857, Naïve Bayes at 0.855, k-Nearest Neighbor at 0.843, and Logistic Regression at 0.821, with small gaps between training and test performance that suggest the models were capturing genuine biological signal rather than memorizing noise. Crucially, this discrimination survived the move to entirely external data. In the 60-patient external validation cohort, the Support Vector Machine led with an AUC of 0.839, Logistic Regression reached 0.835, k-Nearest Neighbor 0.830, and both Naïve Bayes and Random Forest achieved 0.817. Calibration curves showed close agreement between predicted and observed probabilities across the model families, indicating that the classifiers were not only separating stable from vulnerable plaques but doing so with probabilities a clinician could reasonably act upon.

The deep learning arm told a complementary story. The authors trained a multilayer perceptron, a from-scratch convolutional neural network, and three ImageNet-pretrained transfer-learning backbones—VGG16, VGG19, and ResNet50—on the annotated regions of interest. Transfer learning, in which networks pre-trained on millions of natural images are fine-tuned on medical data, proved advantageous: VGG16 achieved the best test AUC of 0.767, ResNet50 followed at 0.754, and VGG19 reached 0.718, while the multilayer perceptron and the from-scratch CNN trailed at 0.669 and 0.654 respectively. External validation reproduced this ordering almost exactly, with VGG16 at 0.776, ResNet50 at 0.751, VGG19 at 0.719, and the simpler architectures in the high-0.68 range. Although the deep models did not surpass the radiomics classifiers in raw discrimination, their performance held up across centers, and they brought a distinct advantage: interpretability through attention mapping.

Using Gradient-weighted Class Activation Mapping, or Grad-CAM, the team visualized which parts of each image the networks attended to when making their predictions. The resulting heatmaps localized hyperattenuating regions—denser, brighter areas within the plaque that correspond to features such as calcification or hemorrhage—and these localizations were consistent with the pathological findings, most clearly for VGG16 and ResNet50. This alignment between machine attention and histological ground truth is more than a technical curiosity. It provides a sanity check that the networks are not exploiting scanner-specific artifacts or incidental image features, but are genuinely looking at the plaque biology that pathologists confirm under the microscope. For a field often criticized for black-box opacity, such visual evidence of anatomically plausible reasoning is a meaningful step toward clinical trust.

The clinical implications are considerable. Current decision-making for carotid stenosis leans heavily on the degree of luminal narrowing, yet the literature has long shown that plaque composition and stability, not stenosis alone, determine rupture risk. A patient with moderate narrowing but a vulnerable plaque may face a higher stroke risk than a patient with severe narrowing and a stable lesion. A validated, automated tool that reads a routine CT angiography scan—the same scan already obtained during standard preoperative workup—and reports a calibrated probability of plaque vulnerability could therefore reshape preoperative risk stratification, help prioritize patients for carotid endarterectomy or stenting, and guide the intensity of medical therapy and surveillance for those managed conservatively. Because the radiomics models in this study used conventional machine learning classifiers on compact feature sets, they are also lightweight and deployable, requiring no exotic hardware and lending themselves to integration into existing picture archiving and communication systems.

The authors are appropriately measured about what their framework can and cannot yet do. The study is retrospective, and the deep learning models, while externally validated, did not reach the discrimination levels of the best radiomics classifiers, suggesting that hand-crafted quantitative descriptors of plaque texture and morphology still carry information that current convolutional architectures do not fully capture from cropped regions of interest. The natural next steps include prospective validation in consecutive clinical cohorts, testing across scanner vendors and imaging protocols, and exploration of hybrid models that fuse radiomic features with deep-learned representations. Funding for the work came from the Shaoxing City Science and Technology Plan Project 2023, and the study was conducted under the Declaration of Helsinki with institutional ethics approval and informed consent from all participants. Even with those caveats, the convergence of pathology-validated labels, multi-center external testing, strong calibration, and interpretable attention maps makes this one of the more convincing demonstrations that AI-assisted plaque stability assessment is moving from proof of concept toward a genuinely usable instrument in the fight against ischemic stroke.

Subject of Research: AI-based carotid plaque stability assessment on CT angiography for ischemic stroke risk prediction

Article Title: Integrated radiomics and image-based deep learning framework using carotid CT angiography ROI datasets for plaque stability assessment as a predictor of ischemic stroke risk

Article References: Lu, M., Xu, Z., Hanfei, T., Luo, P., Huang, F., Wang, G., Bi, L., Jiang, N., Lang, D., Daqiao, G., & Wang, G. (2026). Integrated radiomics and image-based deep learning framework using carotid CT angiography ROI datasets for plaque stability assessment as a predictor of ischemic stroke risk. BioMedical Engineering OnLine. https://doi.org/10.1186/s12938-026-01621-7

Image Credits: AI Generated

DOI: 10.1186/s12938-026-01621-7

Keywords: carotid CT angiography, radiomics, deep learning, plaque stability, ischemic stroke, stroke risk stratification, carotid endarterectomy, machine learning, VGG16, ResNet50, Grad-CAM, biomedical engineering

The correction carries its own digital object identifier, 10.1007/s40846-026-01048-4, permanently anchoring the notice to the scholarly record, while the underlying research remains citable at 10.1007/s40846-024-00922-3 as volume 44, pages 799 to 808, of the journal. Springer’s version of record for the correction is dated 27 August 2026, and the document participates in Crossmark, the cross-publisher initiative that flags readers whenever a paper they are viewing has been updated. What the notice does not do is explain how the error arose. It does not say whether the wrong image file was uploaded during production, whether a panel was mislabeled, or whether the mistake was caught by the authors, a reader or the editorial office. It simply presents the correct figure and confirms that the original article has been corrected. Typically rendered as a ranked bar chart, the figure shows at a glance which variables the model leans on most — precisely why its accuracy matters.

Behind the notice stands a research team that spans two complementary sides of the neurovascular problem. Mariyam Argymbay, Shams Khan, Noman Ahmad and Yasin Mamatjan are based in the Faculty of Science at Thompson Rivers University in Kamloops, British Columbia, with Mamatjan serving as corresponding author. Mira Salih is affiliated with the Brain Aneurysm Institute at Harvard Medical School and Beth Israel Deaconess Medical Center in Boston, a clinical environment devoted to the vascular pathologies that can precipitate devastating brain events. The pairing is telling. Stroke risk assessment is not purely a software exercise; it demands fluency in the epidemiology of hypertension, atrial fibrillation, diabetes and the other conditions that precede cerebrovascular accidents, and it demands a sense of how probabilistic information lands on an actual patient. A collaboration that joins a Canadian computing and biomedical engineering group with a Harvard-affiliated aneurysm research institute is exactly the kind of coalition this problem tends to attract.

The system the team describes is, at its core, a machine-learning pipeline wearing two hats. The first hat is predictive. Like clinical risk models before it, a recommender system for stroke risk assessment ingests patient variables — the kinds of features that dominate stroke epidemiology, such as age, blood pressure, diabetes status, cardiac rhythm abnormalities, smoking history and prior vascular events — and produces an estimate of an individual’s probability of stroke. Systems of this type are usually validated retrospectively, trained and tested on recorded patient data with performance summarized by standard metrics, before anyone contemplates prospective use. The second hat is prescriptive. Where classical risk scores stop at a number, a recommender maps that number onto actions: which screenings, interventions or lifestyle changes are most relevant for a person at a given level of risk. In engineering terms, the recommendation layer is a decision-support component that converts a calibrated probability into prioritized, personalized guidance — conceptually closer to how streaming platforms convert viewing histories into watchlists, except the stakes are measured in neurons rather than evenings.

The Strokebot is the conversational face of that machinery — a chatbot integrated directly into the risk-assessment workflow rather than bolted on afterward. Health chatbots of this kind typically conduct structured dialogue to gather or confirm risk-relevant information, translate an abstract risk score into plain language, answer follow-up questions and steer users toward appropriate care, including education about the sudden facial drooping, arm weakness and speech difficulty that mark stroke’s warning signs. The design logic is friction reduction. A risk model locked behind a dashboard helps experts; a risk model that talks helps everyone else. Integration also matters for data flow, because a conversational agent that feeds the underlying recommender can, in principle, keep the model’s inputs current and its recommendations aligned with what the user has actually been told. No credible chatbot claims diagnostic authority; the goal is triage and engagement rather than replacement of physicians, and responsible implementations keep a human clinician firmly in the loop.

The corrected Figure 3 concerns the system’s third role, and arguably its most important one: self-explanation. SHAP — SHapley Additive exPlanations — imports a concept from cooperative game theory devised by economist Lloyd Shapley in the 1950s, work later honored with a Nobel Memorial Prize. Shapley’s question was how to divide a game’s payout fairly among players whose contributions differ. SHAP recasts a machine-learning prediction as exactly that game: each input feature is a player, the prediction is the payout, and a feature’s Shapley value is its average marginal contribution to the prediction, computed across all possible orderings of the players. The result is additive and locally faithful — the prediction equals a baseline value plus the sum of every feature’s contribution — which is why SHAP has become one of the most widely used tools for opening up otherwise opaque models such as gradient-boosted tree ensembles and neural networks. Exact Shapley computation grows combinatorially with feature count, so practical implementations rely on model-structure shortcuts and careful sampling to make the arithmetic tractable at real-world scale.

When Shapley values are computed for every individual in a dataset, their absolute magnitudes can be averaged into a single global picture of what the model relies on most. That averaged, ranked summary is what Figure 3 presents: a SHAP-based global risk factor ranking showing which inputs the stroke model weights most heavily across the population it learned from. For clinicians, such a chart functions as a contract. If the model promotes a biologically implausible factor to the top, or buries blood pressure beneath noise variables, the discrepancy is a red flag visible before the system ever reaches a patient. If the ranking instead tracks established stroke epidemiology, it builds confidence that the algorithm has learned medicine rather than artifacts. This is why an incorrectly published ranking figure is not a cosmetic problem. It is a misdelivery of the model’s most consequential self-description, read by anyone skimming the paper for the one picture that summarizes a thousand lines of code.

The timeline is also instructive. Roughly twenty months separate the original publication in December 2024 from the correction in August 2026, an interval that reflects the ordinary rhythms of post-publication scrutiny rather than scandal. Corrections are among the most common documents in scientific publishing, and the infrastructure surrounding them — persistent identifiers, Crossmark badges, version-of-record timestamps — exists precisely so that an updated figure can supersede a faulty one without erasing the historical trail. The original article’s page now leads readers to the corrected version, preserving the citation trail while ensuring the fixed figure is what most visitors encounter. The alternative, silently swapping an image inside a published paper, would corrode the very trust that identifiers and archives are built to protect. In fast-moving fields where machine-learning health papers accumulate citations quickly, a DOI-anchored correction ensures that anyone citing, reproducing or deploying the work meets the amended version first. The machinery worked as designed: slowly, visibly and on the record.

The broader stakes are difficult to overstate. Stroke remains one of the world’s leading causes of death and long-term disability, and widely cited global estimates put new cases at well over ten million each year, with projections suggesting the burden will climb as populations age. The encouraging corollary, reinforced by decades of epidemiological research, is that the large majority of stroke risk is tied to detectable, modifiable factors — with elevated blood pressure consistently emerging as the single most powerful one — which is why tools that can find at-risk individuals early and talk them toward prevention hold such appeal for strained health systems. Global prevention campaigns have drilled the same message for years: control hypertension, treat atrial fibrillation with anticoagulation where indicated, manage diabetes and cholesterol, quit smoking, keep moving. An explainable model that reproduces those priorities and personalizes them to an individual’s profile could extend their reach. But deployment hinges on credibility, and credibility requires that the model’s published explanation be exactly what its authors intended.

Figure 3 now reads as its authors intended, and a correction notice of a few hundred words has quietly done its job. The episode is a useful reminder that in medical artificial intelligence, the explanation is part of the intervention. A Strokebot can only be as trustworthy as the risk model beneath it, and the risk model can only be as trustworthy as the published evidence of how it weighs the world. When that evidence appears in error, the whole chain of trust wobbles; when it is corrected, one link at a time and on the record, the chain holds. Science’s smallest genre, the erratum, rarely goes viral. But it is where the discipline does its most honest bookkeeping — and in this case, it is where a machine’s account of stroke risk was set right.