The majority of patients diagnosed with pancreatic cancer have disease that is already too advanced for surgical removal at the time of diagnosis. For these patients with unresectable pancreatic cancer, treatment options are limited and largely palliative in intent, making prognosis prediction particularly important for guiding care decisions.
Interstitial brachytherapy is a localized radiation therapy approach in which radioactive seeds are implanted directly into or near the tumor under image guidance. This technique delivers high radiation doses to the tumor while minimizing exposure to surrounding normal tissue, and has been explored as a palliative or disease-control strategy in patients who are not surgical candidates.
Monitoring the response of a pancreatic tumor to brachytherapy is challenging because standard imaging often cannot detect subtle early changes in tumor viability. A more sensitive imaging-based monitoring tool could help clinicians identify responders from non-responders early, allowing timely adjustment of the treatment plan.
Endoscopic ultrasonography (EUS) is a minimally invasive imaging technique that provides high-resolution images of the pancreas from within the gastrointestinal tract. EUS is particularly useful for visualizing pancreatic tumors and guiding interventional procedures, making it a natural platform for developing image-based treatment response monitoring tools.
The study enrolled 25 patients with unresectable pancreatic cancer who were treated with interstitial brachytherapy. Serial EUS imaging was performed before and after brachytherapy, providing paired images that could be compared to detect treatment-induced changes in tumor texture and appearance.
EUS images were analyzed using a novel image processing algorithm based on fuzzy logic classification. Fuzzy classification is particularly appropriate for medical image analysis because it handles the inherent uncertainty and gradation in tissue appearance rather than imposing rigid binary boundaries between tissue categories.
The algorithm extracted quantitative texture features from EUS images of the tumor before and after brachytherapy. The change in the fuzzy classification score between the pre-treatment and post-treatment EUS images was used as a quantitative index of treatment response, reflecting shifts in tissue echogenicity and heterogeneity caused by radiation-induced tumor changes.
Overall survival (OS) was the primary clinical outcome measure. Statistical correlation analysis was performed to test whether the change in fuzzy classification score following brachytherapy correlated with the length of patient survival, establishing the index as a potential prognostic marker.
Fuzzy logic is a mathematical framework that allows variables to take on degrees of membership in multiple categories simultaneously, rather than being assigned to a single discrete category. In image analysis, this means that a pixel or region can be partially classified as belonging to different tissue types, reflecting the genuine ambiguity of biological tissue boundaries in ultrasound images.
The fuzzy classification algorithm analyzes the statistical distribution of pixel intensities within the tumor region of interest in the EUS image. It computes membership values across multiple predefined echogenicity classes, and the composite score summarizes the overall textural state of the tumor at a given time point.
Ultrasound image texture is directly related to the acoustic properties of the tissue being imaged, which in turn reflect the tissue's biological composition. Tumor necrosis, fibrosis, vascular changes, and cellular density all influence the EUS appearance, making texture a proxy for treatment-induced biological changes within the tumor mass.
The change in fuzzy score from baseline to post-treatment provides a longitudinal measure of how much the tumor's textural characteristics have shifted in response to brachytherapy. A decrease in the score was hypothesized to indicate favorable tissue response, and this hypothesis was tested against observed survival outcomes.
The primary finding was a significant positive correlation between the change in fuzzy classification score and overall survival, with a Pearson correlation coefficient r = 0.616 and statistical significance at P = 0.001. This moderate-to-strong correlation indicates that the EUS-derived image metric carries substantial prognostic information.
Patients whose fuzzy classification score decreased after brachytherapy, indicating a shift toward a different tissue composition pattern, had a median overall survival of 151 days. In contrast, patients whose score did not decrease survived a median of only 67 days. This difference represents a 2.25-fold improvement in median survival in the responding group.
The survival difference between groups with decreased versus non-decreased fuzzy scores was statistically significant, confirming that the image-processing metric captures clinically meaningful prognostic information beyond standard clinical staging variables alone.
These results suggest that the fuzzy score change is not merely a statistical artifact but reflects genuine biological differences in how individual tumors respond to brachytherapy, differences that are visibly encoded in the EUS image texture and detectable by the automated processing algorithm.
The ability to identify brachytherapy responders from non-responders using an objective imaging metric has direct clinical value. For patients who show early evidence of favorable tumor response by EUS texture analysis, clinicians can have greater confidence in continuing the current treatment strategy and can focus supportive care planning around a longer anticipated survival period.
For patients who show no evidence of favorable texture change, early identification of treatment failure allows clinicians to consider alternative interventions, including systemic chemotherapy, pain management optimization, or enrollment in clinical trials, rather than continuing with a therapy unlikely to provide benefit.
The use of EUS is particularly advantageous because EUS-guided brachytherapy placement is already part of the clinical workflow for many patients with unresectable pancreatic cancer. Adding the image texture analysis step to existing EUS examinations would require minimal additional burden to patients or clinical infrastructure.
The relatively small sample size of 25 patients is a limitation, but the strength of the correlation and the magnitude of the survival difference between groups provide compelling preliminary evidence that warrants validation in larger prospective studies before clinical adoption.
This study establishes a novel application of fuzzy logic image processing to EUS images as a quantitative prognostic tool for unresectable pancreatic cancer treated with brachytherapy. The significant correlation between texture score changes and survival demonstrates that meaningful biological information is encoded in the EUS image beyond what visual inspection alone can reveal.
The approach is conceptually simple, computationally efficient, and built on a widely available imaging platform, suggesting favorable feasibility characteristics for future clinical implementation. The automated nature of the algorithm also reduces the subjectivity inherent in qualitative visual assessment of treatment response on EUS.
Future work should validate the method in larger multicenter cohorts, compare the prognostic value of EUS texture analysis to other response monitoring modalities such as CT volumetry or PET-CT, and explore whether the algorithm can be adapted to predict response to other treatment modalities such as ablation or chemotherapy.
In the broader context of pancreatic cancer care, the ability to predict prognosis early in the treatment course using noninvasive or minimally invasive imaging tools could facilitate more individualized patient management, ensuring that each patient receives the care approach most likely to benefit their specific tumor biology and treatment response profile.