Emerging non-invasive imaging biomarkers of Ki-67 in pancreatic cancer: Toward predictive precision oncology

Abstract

The proliferative index of Ki-67 in pancreatic ductal adenocarcinoma (PDAC) exhibits strong correlations with tumor progression and prognosis, holding significant clinical implications. Yang et al employed contrast-enhanced ultrasound (CEUS) to indirectly evaluate neovascularization in pancreatic cancer lesions. Specific CEUS parameters demonstrated significant diagnostic value in assessing Ki-67 expression. The falling slope 50% achieved an area under the curve of 0.838. Meanwhile, the rise slope 10%-90% exhibited superior overall diagnostic accuracy (area under the curve = 0.863), showing a sensitivity of 0.92 and a moderate specificity of 0.759. These values demonstrate specificity in differentiating between low and high Ki-67 expression groups. This study effectively addresses the critical need for a non-invasive assessment of pancreatic cancer aggressiveness via Ki-67 expression. These findings strongly support the translational potential of CEUS biomarkers for non-invasive Ki-67 assessment and treatment stratification in PDAC. While Yang et al demonstrated exhibited encouraging methodologies, its retrospective design, modest sample size, and single-center nature may impede generalizability, pending validation in multi-institutional cohorts. We recommend expanding the sample size to enhance representativeness and adopting prospective studies integrating multimodal imaging techniques, such as magnetic resonance imaging and positron emission tomography to improve diagnostic reliability. This study is the first to integrate insights from CEUS, magnetic resonance imaging, and positron emission tomography for Ki-67 expression assessment in PDAC. Building on this innovation, we focus this article on recent advances in the clinical diagnosis of pancreatic cancer, aiming to provide insights for advancing research in this field.

Key Words: Ki-67 expression; Pancreatic cancer; Contrast-enhanced ultrasound; Imaging techniques; Clinical diagnosis

Core Tip: Ki-67 expression levels in tumor cells function as both an early predictor of treatment response and a long-term prognostic factor in cancer patients. In this study, we begin by assessing the limitations of current non-invasive imaging techniques for evaluating Ki-67 in pancreatic cancer. Subsequently, we explore more rational methodological refinements, including prospective studies and combined imaging modalities, with the objective of enhancing predictive accuracy. This discussion also highlights potential future research directions.

INTRODUCTION

Given the prognostic significance of Ki-67 in pancreatic cancer and the limitations of current biopsy-based evaluations, recent imaging-based methods deserve critical discussion. We critically discuss and advocate for the development and clinical integration of non-invasive imaging biomarkers, notably contrast-enhanced ultrasound (CEUS)-derived perfusion parameters, as predictors of Ki-67 expression capable of guiding therapy. Pancreatic cancer is an aggressive malignancy that ranks as the seventh leading cause of cancer-related mortality worldwide[1]. According to GLOBOCAN estimates, approximately 511000 new cases and 467000 deaths occurred globally in 2022[2], underscoring its extremely poor prognosis. Pancreatic ductal adenocarcinoma (PDAC) is the most common and aggressive form of pancreatic malignancy, accounting for about 90% of cases, exhibiting a five-year survival rate below 10%[3-5] and presenting a significant challenge to global health.

Ki-67 is highly upregulated in most malignant cells[6]. Its expression level inversely correlates with tissue differentiation and shows significant associations with tumor metastasis and clinical stage[7], the Ki-67 expression is a well-established prognostic marker across a wide range of malignancies[8,9] and is of particular prognostic significance in PDAC[10]. Notably, recent studies have demonstrated that elevated Ki-67 expression in PDAC directly correlates with significantly reduced patient survival, thereby establishing it as a critical independent prognostic factor[10]. Evidence indicates that modulating Ki-67 activity or expression can interfere with tumorigenesis and progression, highlighting its potential as a therapeutic target in cancer[7,11]. This finding holds considerable implications for the development of novel therapeutic avenues for PDAC. However, the current Ki-67 assessment in PDAC research remains methodologically constrained[12] by its reliance on invasive surgical or biopsy specimens, limiting applicability in patients without tissue access and precluding longitudinal monitoring. Addressing this clinical imperative, our study aims to develop non-invasive Ki-67 quantification techniques [CEUS, magnetic resonance imaging (MRI) and positron emission tomography (PET)] and systematically evaluate their comparative merits, limitations, and multimodal integration potential[13-15].

Current evidence

Yang et al[16] collected pancreatic mass tissue samples from 54 PDAC patients at the First Affiliated Hospital of China Medical University. Ultrasound (US) was performed using Siemens ACUSON Sequoia (Germany) and Mindray Resona R9 Exp (China) systems with convex abdominal probes (1-5 MHz). The contrast agent SonoVue (Bracco, Italy) was mixed with 5 mL of normal saline. After conventional US localized the pancreatic lesions, contrast mode was activated. A 2.4-mL bolus was injected rapidly (2-3 seconds) into a superficial cubital vein, followed immediately by a 5-mL saline flush; timing began upon injection. Lesion enhancement and washout patterns were observed in real-time with probe stability maintained (> 2 minutes); images were archived. The authors calculated the Ki-67 expression by assessing the proportion of Ki-67-positive tumor cell nuclei after counting 1000 tumor cells. Using 50% as the threshold, samples were stratified into high and low Ki-67 expression groups. Subsequently, the authors employed CEUS to quantitatively evaluate perfusion-related parameters in tumor tissue, aiming to develop a non-invasive prediction model for evaluating Ki-67 expression and pathological staging in pancreatic cancer. No statistically significant differences (P > 0.05) were observed between the high and low Ki-67 expression groups regarding age, sex, carbohydrate antigen 19-9 levels, tumor location, differentiation degree, and clinical stage, indicating comparable baseline characteristics.

The authors subsequently performed Spearman rank correlation analysis to assess the correlations between Ki-67 index and quantitative CEUS parameters. These findings revealed highly significant positive correlations between rise slope 10%-90% (Rs1090) and Ki-67 expression (P < 0.001). Conversely, falling slope 50% (Fs50) exhibited a highly significant negative correlation with Ki-67 expression (P < 0.001). Receiver operating characteristic curve analysis evaluated the diagnostic efficacy of individual CEUS parameters in differentiating between high and low Ki-67 expression groups. Yang et al[16] report that the Fs50 yielded the highest area under the curve (AUC) of 0.838, demonstrating its superior diagnostic accuracy for low Ki-67 expression. The parameters rise slope 50%, maximum intensity (IMAX), wash-out rate and Rs1090 all achieved AUC values > 0.75, indicating strong diagnostic performance for high Ki-67 expression. This retrospective study of 54 samples provided initial evidence supporting the feasibility of using quantitative CEUS parameters for non-invasive assessment of Ki-67 expression in pancreatic cancer. These findings corroborated the potential of CEUS-derived perfusion parameters as non-invasive biomarkers for assessing tumor angiogenesis and predicting proliferative activity in pancreatic cancer, thereby offering imaging-based support for the optimization of personalized anti-angiogenic and proliferation-targeting therapeutic strategies.

Using well-defined quantitative CEUS parameters and rigorous statistical analyses, including receiver operating characteristic curves, this study objectively evaluated the predictive performance of Ki-67 levels. These analyses provide foundational evidence for developing noninvasive imaging biomarker models to assess Ki-67 expression and pathological staging in pancreatic cancer, directly addressing the urgent need for methods of evaluating tumor aggressiveness. To further strengthen the generalizability and robustness of these promising findings, future work must prioritize rigorous external validation of the identified CEUS parameters (e.g., Rs1090, Fs50) across diverse, independent patient cohorts. While multi-center prospective studies integrating multimodal algorithms are crucial, the immediate need is to confirm the diagnostic performance and generalizability of these specific biomarkers beyond the initial single-center setting. The current lack of external validation represents a significant limitation for clinical translation. Addressing this requires concerted efforts, potentially through international consortia (e.g., dedicated pancreatic imaging biomarker groups) or open data-sharing initiatives, to pool larger datasets and enable robust cross-validation. Here are more details.

Firstly, the retrospective analysis only included 54 patient samples. As acknowledged by the authors, the limited sample size and single-center origin risk selection bias, as well as substantial intergroup variability, are limitations of the study. This potential limitation may introduce bias, as the studied sample does not accurately represent regional diversity. Future studies could expand cohorts to encompass broader geographic, ethnic, and age distributions, while fostering multi-center collaborations to enhance data robustness. To enhance stability and accuracy, larger sample sizes with stratified sampling based on defined criteria are necessary[17,18].

Secondly, we note that the retrospective design relies on pre-existing medical records, introducing vulnerability to selection bias that may constrain conclusions. Before clinical translation, prospective validation is required to better control experimental conditions, minimize bias, and enhance reliability[19].

Thirdly, integrating multiple imaging modalities offers substantial value for comprehensive lesion characterization. MRI provides excellent visualization of pancreatic anatomy, clearly delineates peri-pancreatic lymph nodes, and sensitively detects hepatic metastases, offering superior soft-tissue contrast for early tumor identification[20,21]. PET is a medical imaging procedure that utilizes radiolabeled glucose analogs, which accumulate preferentially in metabolically active tumors. The resulting tracer uptake kinetics allow PET to complement MRI in characterizing indeterminate lesions, identifying extra-pancreatic metastases, quantifying systemic tumor burden, and reliably distinguishing malignancy from chronic pancreatitis[22]. CEUS provides unique advantages in real-time visualization of tumor perfusion dynamics, enabling quantitative assessment of functional neovascularization characteristics (microvessel density) through hemodynamic parameter analysis. These metrics reflect microvascular functional integrity rather than structural anatomy[16]. Conversely, PET excels at evaluating metabolic activity in tumor cells, with specific tracers capable of assessing metabolic reprogramming in vascular endothelial cells[23]. Together, CEUS and PET comprehensively characterize tumor angiogenesis: CEUS deciphers hemodynamic phenotypes (perfusion kinetics), while PET reveals metabolic adaptations (substrate utilization). Integrating CEUS perfusion kinetics (Rs1090 AUC = 0.86) with MRI diffusion metrics [apparent diffusion coefficient (ADC) AUC = 0.83] and maximum standardized uptake value (SUVmax) of PET metabolic activity (SUVmax AUC = 0.81) yields synergistic diagnostic gain, achieving a pooled AUC of 0.92 (95% confidence interval: 0.89-0.95) in multi-modal Ki-67 prediction models[24,25]. Building upon CEUS foundations, the integration of complementary modalities like MRI and PET may overcome inherent technical constraints. Novel fibroblast activation protein inhibitor PET tracers show enhanced specificity for Ki-67-rich tumors with desmoplastic stroma, overcoming limitations of FDG-PET in hypometabolic yet proliferative lesions[26,27]. By targeting cancer-associated fibroblasts prevalent in high-Ki-67 microenvironments, fibroblast activation protein inhibitor PET improves tumor conspicuity and prognostic stratification for stromal-dominant malignancies.

Fourthly, in this study, the methodology of counting 1000 tumor cells to determine Ki-67 positivity percentage, using a 50% expression cutoff, aligns with international standards and mitigates sampling error[12]. However, studies indicate that the intra-class correlation coefficient remains significant during manual counting of 1000 cells, particularly near the cutoff value[28,29]. While manual counting of 1000 cells remains the conventional gold standard for Ki-67 quantification, its reliability is constrained by inter-observer variability (intra-class correlation coefficient: 0.45-0.62) and selection bias in hotspot identification. To mitigate these limitations, we advocate adopting automated whole-slide imaging coupled with artificial intelligence (AI) validation as a methodological imperative. AI algorithms achieve > 90% concordance with expert consensus by standardizing tumor-region segmentation and eliminating counterstain interference, thereby establishing a new benchmark for reproducible proliferation assessment[28,30,31].

Additionally, CEUS interpretation exhibits significant inter-observer variability, primarily attributable to operator-dependent acquisition and analysis techniques. To mitigate operator subjectivity, AI-guided interpretation systems demonstrate compelling efficacy[32]. Furthermore, research has demonstrated that AI can perform classification tasks for focal lesions[33-35] and predict Ki-67 expression[24,36], including distinguishing between benign and malignant lesions or categorizing lesion subtypes[37]. Notably, these AI-assisted strategies achieve a level of diagnostic accuracy comparable to that of experts when using comprehensive clinical data, while also significantly reducing inter-observer variability in CEUS interpretation[38]. The utilization of AI in the automatic extraction of complex radiomic features from native CEUS, MRI, or PET data facilitates the identification of novel imaging biomarkers. The integration of these multi-parametric features facilitates the development of predictive models for Ki-67 status, pathological grade, and molecular characteristics. In addition, emerging transformer-based architectures now enable deeper integration of imaging and contextual clinical data. Large language models like LLaMA-RAD can process radiology reports to extract semantic features correlating with Ki-67 expression patterns, while vision-language models (e.g., RadFM) fuse image-text embeddings to generate probabilistic Ki-67 maps. When combined with radiomic analysis, these frameworks achieve 12%-18% higher accuracy in predicting high-grade proliferation compared to conventional convolutional neural networks[39,40].

Multimodal integration

Imaging techniques remain pivotal for the initial diagnosis and accurate staging of pancreatic cancer in clinical practice, with non-invasive modalities playing critical roles (Figure 1). These imaging techniques exhibit distinct advantages and limitations in assessing Ki-67 expression. US serves as a primary screening tool due to its accessibility and real-time capability, effectively visualizing pancreatic parenchyma[41,42] and demonstrating efficacy in evaluating Ki-67 proliferation status in PDAC[43]. Computed tomography (CT) enables indirect assessment through texture features, though limited by insufficient soft-tissue contrast[44]. Despite this constraint, CT provides superior spatial-temporal resolution[45-47], facilitating Ki-67 evaluation via intratumoral or peritumoral data[48,49]. MRI provides quantitative characterization of tissue microstructural heterogeneity through texture analysis, whereas it cannot directly measure cellular proliferative activity[50]. The ADC, derived from diffusion-weighted MRI, has been shown to offer quantitative insights into tumor cellularity[51,52]. Notwithstanding this limitation, MRI, including magnetic resonance cholangiopancreatography, provides complementary information to contrast-enhanced CT, detecting tumors, predicting recurrence and vascular invasion, and serving as a prognostic indicator[14,20,53]. Furthermore, imaging analysis of these tissues has proven to be a reliable method for predicting Ki-67 expression[54]. PET sensitively detects metabolic hyperactivity associated with proliferation[26,27], but cannot distinguish proliferation-specific metabolism from background activity, exhibiting limited sensitivity in hypometabolic tumors[23]. PET/CT delineates tumor metabolic activity and burden, excelling in whole-body disease assessment[15,55]. Evidence has shown that PET demonstrates superior accuracy in predicting Ki-67 expression in various tumors[56]. CEUS offers dynamic assessment of microvascular perfusion kinetics with high temporal resolution, capturing specific hemodynamic parameters[16]. Nevertheless, its application is restricted in deep-seated lesions and lacks macrostructural characterization capabilities. It quantifies tumor perfusion, holding significant value for early diagnosis and therapeutic evaluation[57].

Figure 1 Imaging modalities and their contribution to Ki-67 assessment. An integrated clinical pathway for tumor management involves initial screening via ultrasound to assess boundary and rigid features, followed by contrast-enhanced ultrasound biomarker testing. This testing includes quantitative parameters such as rise slope 10%-90%, maximum intensity, and falling slope 50%. Contrast-enhanced ultrasound testing serves as a trigger for multiparametric magnetic resonance imaging, which is used for apparent diffusion coefficient and Ki-67 assessment. Treatment is stratified by Ki-67 expression: High expression mandates neoadjuvant therapy, while low expression proceeds directly to surgical evaluation. The employment of advanced quantification techniques and positron emission tomography metabolic assessment further reinforces the precision oncology decision-making process throughout the pathway. US: Ultrasound; CEUS: Contrast-enhanced ultrasound; Rs1090: Rise slope 10%-90%; IMAX: Maximum intensity; Fs50: Falling slope 50%; MRI: Magnetic resonance imaging; PET: Positron emission tomography.

Among these modalities, US, CT, and CEUS are optimally suited as first-line imaging tools in routine clinical practice, whereas PET/CT and MRI are preferentially reserved for specialized tertiary care settings. While existing clinical imaging modalities each exhibit unique advantages and inherent limitations, the strategic integration of multiple techniques can significantly enhance diagnostic accuracy by leveraging their strengths[41] (Table 1).

Table 1 Advantages and limitations of modalities used in the diagnosis of pancreatic cancer.

Methods	Advantages	Limitations	Clinical utility for Ki-67 estimation	Utility in Ki-67 prediction	Validation status
US	Easy to use, flexible and intuitive, non-invasive and radiation-free	Difficulty visualizing the entire pancreas	No validated biomarkers	Low	Research-only
CT	Excellent spatial and temporal clarity	Inability to diagnose pancreatic lesions accurately	Texture analysis (low entropy excludes high Ki-67 expression)[67]	Moderate	Transitional
MRI	Exceptional imaging for depicting local pancreatic disease	Expensive and less widely available	Diffusion restriction (low ADC) predicts high Ki-67 expression for preoperative risk stratification[48,49]	High	Limited clinical use
PET	Advantages are obvious when detecting extrapancreatic metastasis and evaluating the tumor load throughout the body	Expensive, less widely available and contrast exposure	Focal hypermetabolism (high SUV) triggers targeted biopsy for suspected aggressive lesions[58]	Moderate	Routine clinical
CEUS	The major blood vessels and microvascular system within the pancreas can be displayed	Highly subjective and its accuracy is highly influenced by the operator’s skill level and experience	Elevated perfusion kinetics (high Rs1090) may reduce reliance on biopsy in high-bleeding-risk patients	High	Research-only

Research-only: Preliminary evidence, requires technical optimization. Transitional: Retrospective validation, prospective trials ongoing. Limited clinical: Single-center prospective validation, pending multi-center confirmation. Routine clinical: Prospective multi-center validation, incorporated into major guidelines. US: Ultrasound; CT: Computed tomography; MRI: Magnetic resonance imaging; PET: Positron emission tomography; CEUS: Contrast-enhanced ultrasound; ADC: Apparent diffusion coefficient; SUV: Standard uptake value; Rs1090: Rise slope 10%-90%.

To overcome the inherent limitations of standalone imaging modalities, we propose developing explainable deep learning frameworks that integrate multi-modal radiomic features from PET, MRI, and CEUS. These models should leverage cross-modal attention mechanisms to generate “virtual Ki-67 proliferation maps”, with technical validation focusing on three critical aspects (Figure 2). This integrated approach has the potential to provide more robust diagnostic information, thereby advancing pancreatic cancer imaging and improving patient management.

Figure 2 Unified predictive model for Ki-67 using multimodal imaging. This schematic presents an artificial intelligence fusion model that integrates multimodal imaging data. These include contrast-enhanced ultrasound derived perfusion kinetics, specifically rise slope 10%-90%, positron emission tomography based metabolic activity represented by the maximum standardized uptake value, and magnetic resonance imaging derived diffusion and texture metrics such as apparent diffusion coefficient and entropy. The model extracts key imaging features including time-intensity curves, metabolic volumes, and texture maps, which are then synthesized using a cross-modal attention gate to generate a probabilistic Ki-67 output ranging from 0%-100%. Clinically, this artificial intelligence-derived output supports three major decisions. First, biopsy triage, where a high rise slope 10%-90% value may help avoid invasive procedures in patients at risk for biopsy-related complications. Second, therapy monitoring, where significant changes in perfusion parameters or rising maximum standardized uptake values during neoadjuvant treatment can prompt early adjustments to chemotherapy. Third, prognostic stratification, where high-probability Ki-67 spatial predictions assist in identifying aggressive tumor regions and support surgical margin planning. AI: Artificial intelligence; Rs1090: Rise slope 10%-90%; CEUS: Contrast-enhanced ultrasound; PET: Positron emission tomography; SUVmax: Maximum standardized uptake value; MRI: Magnetic resonance imaging; ADC: Apparent diffusion coefficient.

Future directions

For future clinical translation, multimodal imaging biomarkers that have been validated against histopathological correlates demonstrate compelling potential to strategically supplement or replace invasive biopsies in the management of pancreatic cancer. The incorporation of these noninvasive biomarkers facilitates clinical stratification in patients for whom biopsy is contraindicated. CEUS parameters, such as Rs1090/IMAX, exhibit a strong correlation with Ki-67 expression, thereby enabling non-invasive assessment. Specifically, Rs1090 predicts high Ki-67 expression, supporting its utility in tumor aggressiveness evaluation[16]. Furthermore, the combination of PET/MRI fusion technology has been shown to enhance the correlation between metabolic activity SUVmax by considering structural heterogeneity (entropy). Elevated SUVmax values have been observed to be indicative of increased glycolytic activity, while increased entropy has been found to quantify architectural disorganization, which is often associated with aggressive phenotypes[25,58,59]. This framework facilitates the transition of Ki-67 assessment from static histopathology to dynamic in vivo monitoring, thereby enabling a transition towards dynamic therapeutic monitoring with clinical implications.

To enable cross-site comparability of radiomic features, we advocate for standardized imaging protocols across modalities. CEUS: Uniform contrast administration procedures and dynamic acquisition duration. MRI: Consistent diffusion-weighted sequences and anatomical coverage. PET/CT: Harmonized tracer administration and reconstruction methodologies. Implementation should follow international consensus guidelines (e.g., QIBA, ESUR) to ensure technical reproducibility.

Translating this paradigm into clinical workflows requires a structured algorithm integrating quantitative biomarkers. As illustrated in Figure 3, this algorithm aims to seamlessly integrate quantitative imaging biomarkers into the management paradigm of suspected PDAC cases, particularly in scenarios where core biopsy is challenging or contraindicated. This algorithm utilizes CEUS as a primary source of imaging biomarkers, thereby facilitating integrated analysis through multi-parametric MRI/PET imaging. It aims to utilize imaging biomarkers to provide novel clinical insights for therapeutic strategies.

Figure 3 Integrated imaging biomarker pathway for Ki-67 prediction and treatment allocation in biopsy-contraindicated pancreatic ductal adenocarcinoma. This clinical workflow outlines a decision-making pathway for suspected pancreatic ductal adenocarcinoma, integrating imaging biomarkers to support personalized treatment planning. Following initial contrast-enhanced computed tomography or magnetic resonance imaging for anatomical staging, the feasibility of tissue biopsy is assessed. If biopsy is possible, histopathological Ki-67 grading provides a direct evaluation of tumor proliferative activity. When biopsy is not feasible - such as in patients on anticoagulation therapy or with poor clinical condition - noninvasive contrast-enhanced ultrasound becomes essential. Quantitative contrast-enhanced ultrasound parameters, including elevated rise slope 10%-90% or maximum intensity, are indicative of high Ki-67 expression (≥ 50%), while a low 50% falling slope suggests low Ki-67 expression (< 50%). Optional supplementary imaging, such as magnetic resonance imaging-derived apparent diffusion coefficient or positron emission tomography-derived maximum standardized uptake value, may further enhance risk stratification when needed. These multimodal imaging biomarkers are then integrated via an artificial intelligence-based model to generate a probabilistic Ki-67 score. Based on this prediction, patients are stratified into two therapeutic pathways: Those with high predicted Ki-67 (≥ 50%) are directed toward neoadjuvant therapy due to increased proliferative risk, while those with low predicted Ki-67 (< 50%) are considered suitable for surgical resection planning. PDAC: Pancreatic ductal adenocarcinoma; CT: Computed tomography; MRI: Magnetic resonance imaging; CEUS: Contrast-enhanced ultrasound; Rs1090: Rise slope 10%-90%; IMAX: Maximum intensity; Fs50: Falling slope 50%; ADC: Apparent diffusion coefficient; PET: Positron emission tomography; SUVmax: Maximum standardized uptake value.

The proliferation-imaging stratification framework established for PDAC, which integrates Ki-67-driven therapeutic triage with multiparametric metabolic-vascular phenotyping, demonstrates translational potential for other hyperproliferative malignancies. In neuroendocrine tumors, where Ki-67 expression is a determining factor in World Health Organization grading (G1-G3), CEUS-derived quantitative kinetics (Rs1090/IMAX) may refine prognostic stratification through their established correlation with proliferation activity[60-62]. In this context, longitudinal somatostatin receptor PET imaging has the potential to facilitate dynamic monitoring of treatment response[63,64]. For hepatocellular carcinoma, which exhibits Ki-67-associated microvascular invasion patterns, integrating intravoxel incoherent motion diffusion-weighted imaging quantitatively augments hemodynamic heterogeneity assessment beyond current Barcelona Clinic Liver Cancer staging criteria[65-67]. In neuroendocrine tumors, where Ki-67 dictates World Health Organization grading (G1-G3), CEUS perfusion kinetics (Rs1090/IMAX) could refine prognostic stratification beyond current mitotic counting, particularly for G2 borderline cases. For hepatocellular carcinoma, combining ADC heterogeneity with Ki-67 probability maps may predict microvascular invasion patterns currently undetectable by conventional Barcelona Clinic Liver Cancer staging.

CLINICAL TRANSLATION PATHWAY

A structured implementation pathway progresses through sequential phases: Initial multi-center technical validation studies to establish protocol harmonization, followed by regulatory certification processes for integrated diagnostics and AI algorithms, culminating in tiered clinical deployment within cancer centers with parallel reimbursement integration. Notable challenges include equipment accessibility disparities (particularly PET/MRI availability), evolving AI regulatory compliance requirements, and the need for multidisciplinary workflow coordination. Prioritized adoption in high-risk subgroups where biopsy is contraindicated is recommended to demonstrate initial clinical utility while addressing these barriers.

Open science framework

Creation of FAIR-compliant open repositories with paired imaging-histopathology data is essential. These must standardize multi-modal acquisition protocols, adopt DICOM- whole-slide imaging for digital pathology integration, and implement federated learning to enable global AI collaboration without data transfer, ultimately democratizing biomarker research.

Prospective clinical trial advocacy

We strongly advocate for the immediate initiation of real-time prospective clinical trials that integrate AI-assisted multimodal imaging (CEUS/MRI/PET) to validate non-invasive Ki-67 prediction against critical clinical outcomes. These trials should prioritize three core objectives: Correlating imaging-derived proliferation indices with recurrence-free survival, conducting head-to-head comparisons against histopathologic gold standards, and assessing the economic impact of biopsy reduction, especially in resource-constrained settings where invasive procedures pose significant challenges.

Expanded clinical utility

The proposed multimodal framework transcends initial diagnostic applications to actively inform critical therapeutic decisions across the oncology care continuum. By leveraging dynamic CEUS perfusion kinetics, it enables personalized adaptation of neoadjuvant therapy regimens based on early treatment response patterns. Concurrently, PET/MRI fusion surveillance of metabolic-structural alterations facilitates pre-symptomatic recurrence detection, while integration of Ki-67 probability profiles with molecular subtyping stratifies patients for targeted clinical trials. This tripartite utility, which spans treatment optimization, recurrence monitoring, and precision trial design, collectively elevates the paradigm from passive diagnosis to active therapeutic guidance.

CONCLUSION

Advanced CEUS, as a non-invasive approach, provides valuable prognostic information and an accurate assessment of tumor aggressiveness in pancreatic cancer patients. Yang et al[16] demonstrated a correlation between CEUS-derived quantitative parameters and the Ki-67 expression, revealing the potential of specific parameters to distinguish Ki-67 expression states. Notably, Fs50 showed promise in predicting low Ki-67 expression, while rise slope 50%, IMAX, wash-out rate, and Rs1090 were more accurate in predicting high expression. These findings offer novel avenues and potential tools for clinical diagnostics. This discussion aims to inform new diagnostic strategies and advance the development of imaging biomarkers, ultimately enhancing clinical translation. we prioritize three translational pathways: (1) Implementing multi-center trials for AI-integrated CEUS-MRI-PET fusion workflows; (2) Standardizing perfusion kinetics via radiomics across imaging platforms; and (3) Developing longitudinal biomarkers for dynamic therapy monitoring. Collectively, these initiatives advance our vision of non-invasive precision diagnostics in PDAC, potentially reducing reliance on invasive biopsies with dynamic imaging biomarkers.

ACKNOWLEDGEMENTS

We thank the reviewers for their comments that helped to improve the manuscript.