Accurately diagnosing pancreatic cancer lesions without the use of invasive procedures remains a “significant clinical challenge,” particularly in detecting pancreatic ductal adenocarcinoma (PDAC) and differentiating lesions from benign or borderline tumors.
Results from recent studies suggest that machine learning-based radiomics models may provide a promising non-invasive approach to pancreatic cancer detection, with potential impact on early diagnosis and enhanced care planning and management.
A team of researchers from the Affiliated Hospital of Qingdao University in China conducted this study and published the results. Nursing Management Journal. They evaluated whether radiological features extracted from CT scans could be combined with machine learning algorithms to reliably differentiate pancreatic lesions.
Researchers utilized imaging and clinical data from 640 patients with pathologically confirmed pancreatic tumors, including 450 patients with malignant lesions, 108 patients with borderline lesions, and 82 patients with benign lesions. Data were stratified into training cohort (70%) and validation cohort (30%).
Regional data from arterial and venous phase CT scans were analyzed, and the researchers used least absolute shrinkage selection operator (LASSO) logistic regression to identify 36 important radiological features for the development of machine learning models, including “random forests, logistic regression, support vector machines, and artificial neural networks.”
The researchers also evaluated the role of carbohydrate antigen 19-9 (CA19-9) as a biomarker for pancreatic cancer evaluation and employed an integrated nomogram that combines radiological features and CA19-9 levels to differentiate between PDAC and borderline tumors.
According to the results, all machine learning models showed good performance in identifying the three types of pancreatic lesions, with the random forest model achieving the best performance data, achieving an area under the curve (AUC) of 0.99 and 0.95 “on the training and validation sets,” respectively. The researchers also highlighted that the model identified CA19-9 as an independent diagnostic factor for PDAC, reinforcing its ongoing clinical relevance.
The study demonstrated that a radiomics-based machine learning model “effectively distinguished between benign, borderline, and malignant pancreatic tumors.” Furthermore, the researchers concluded that “the nomogram combining radiomics features and CA19-9 showed strong performance and great potential to streamline the diagnostic process and facilitate timely care planning for patients suspected of having pancreatic cancer.”
