Introduction
Single-molecule fluorescence microscopy has been researched for diagnosis of disease and detecting drugs or substances at the molecular level. It is known to detect the target proteins more precisely than ELISA and has a higher multiple detection rate, with small sample volume and at low cost. Pancreatic cancer often presents at an advanced stage. So there is a demand for early detection.
Aims & Methods
The aim of this pilot study is to check the feasibility of this technique to diagnose the pancreatic cancer. From April 2016 to June 2021, 16 pancreatic cancer patients and 21 non-pancreatic cancer patients were enrolled in the study at Seoul National University of Bundang Hospital. 15 markers (CA19-9, CRP, CA-125, D-dimer, Lumican, Cathepsin-D, IL-6, LRG-1, IFN-r, Beta-2, CEA, Clusterin, TTR, IL-4, and C5a) were detected from blood sample using single-molecule fluorescence microscopy. To create a train and test set for our research, we split the enrolled patients into halves. The primary outcome of the study is the accuracy, which is defined as the number of correct predictions, where patients are predicted as patients and controls are predicted as controls, divided by the total count. We analyzed the classification accuracy of each group and examined the accuracy based on the combination and number of markers used. Although the sizes of the training set and the testing set were equal, we randomly split each sample into either the training or testing set, and repeated this process 1,000 times.
Results
Eight out of 16 pancreatic cancer patients were used for training and the remaining 8 for testing. For the non-pancreatic cancer patients, 11 of the 21 blood samples were used for training, and the other 10 for testing.
In the classification accuracy of diagnosis, we tested the accuracy by utilizing Gaussian Naïve Bayes and Support Vector Machine. In case of Gaussian Naive Bayes for the train set, the average accuracy was 99.2%, the standard deviation was 0.019. When applied to the test set, the accuracy was 86.3%, with a standard deviation of 0.074. In case of Support Vector Machine for the train set, the average accuracy was 75.6%, the standard deviation was 0.004. When tested on the test set, the accuracy was 77.6% with a standard deviation of 0.076.
For the accuracy according to the number of biomarkers used, the accuracy was calculated by adding one biomarker at a time in the order previously listed. Accuracy tended to increase as markers were added, the highest accuracy was achieved when using 10 biomarkers (88.4%), and adding more biomarkers from the 11th to 15th actually decreased the accuracy (lowest 85.5% to highest 87.8%).
Conclusion
In this pilot study, we identified that single molecule counting technology showed a feasibility for the new technique to differentiate the patients with pancreatic cancer from healthy control. Further studies with more samples and appropriate selection of biomarkers, and optimization of classification model are necessary to fit the best performance of this novel technique. Utilizing this novel technique with multiple biomarkers in combination is expect to enable low-cost, small-sample volume, and high- accuracy diagnosis in pancreatic cancer.