Introduction
Contrast-enhanced harmonic endoscopic ultrasound (CEH-EUS) can demonstrate microvascularization of the tumors. However, most images acquired from CEH-EUS are interpreted subjectively without quantification by time-intensity curves (TICs) analysis.
Aims & Methods
In this study, we used handcrafted software to quantify CEH-EUS images for objective interpretation of tumor microvascularization to make differential diagnosis of pancreatic tumors feasible.
We analyzed the videos of CEH-EUS for pancreatic tumors from the patients in National Taiwan University Hospital and Far Eastern Memorial Hospital. The convex-type echoendoscope (GF-UCT260; Olympus, Tokyo, Japan) equipped with ultrasound (US) systems (EU-ME2) was used for the procedures. The second generation of US contrast agent with Sonazoid (GE Healthcare, Milwaukee,Wisc), which contains perfluorobutane microbubbles with a median diameter of 2–3 μm, was administered bolus and intravenously with a dosage of 0.015 ml encapsulated gas per kilogram body weight. The videos were recorded in AVI formats with 30 frames/second.
The algorithm for getting TICs includes the following steps:
(1) Circling the region of interest(ROI) in the B-mode
(2) Performing speckle tracking of the ROI in the B-mode
(3) Simultaneously obtaining the intensity of ROI (the mean pixel values in the ROI) in the corresponding contrast-enhanced harmonic image
(4) Outputting TICs after subtracting the baseline intensity (defined as mean TIC in the first 10 s for eliminating the different baseline intensities caused by other settings).
(5) Outputting histogram of pixel values and excel file containing all pixel values from the frame with the highest mean pixel value automatically.
(6) Calculating the standard variance of (5) to represent the heterogeneity inside the ROI.
The speckling tracking composes of searching and matching processes. We used full search for the searching process and normalization cross-correlation for the matching process. By using the hardware of CPU with Intel i7-8750H and GPU with Nvidia GTX 1050 Ti, the computation time was 0.7 ms/frame.
Results
The detailed results were shown in Table 1. We collected 10 and 7 videos of CEH-EUS for pancreatic tumors from NTUH and FEMH respectively in this study. Among them, 9 were diagnosed of pancreatic ductal adenocarcinoma (PDAC), 6 were pancreatic neuroendocrine tumors (PNET), and 2 were autoimmune pancreatitis (AIP). The TICs were obtained through our proposed model. Briefly, there were no significant differences in times to peak, the mean peak intensity, the in-slope tangent and standard deviation of pixel value in ROI (from the frame with the highest mean pixel value) among PDAC, PNET, and AIP. However, there was significant difference in the mean peak intensities of PDAC and non-PDAC (37.88 vs 63.12, p=.031).
Table 1.The results from three kinds of pancreatic tumors based on the proposed model
| PDAC
| PNET
| AIP
| Non-PDAC(NET & AIP)
| P value(PDAC vs PNET vs AIP)
| P value
(PDAC vs non-PDAC)
|
Number
| 9 | 6 | 2 | 8 |
|
|
Times to peak(SD), seconds
| 27.50(7.98)
| 26.76(8.56)
| 21.5(5.94)
| 25.45(7.65)
| 0.632
| 0.598 |
The mean peak intensity(SD), dB
| 37.88(18.58)
| 66.10(28.91) | 54.2(3.96)
| 63.12(25.09)
| 0.087
| 0.031 |
In slope(SD)
| 2.90(2.34)
| 5.07(3.87)
| 5.34(2.41)
| 5.14(3.40)
| 0.331
| 0.131 |
SD of pixel value in ROI (the highest mean pixel value)(SD)
| 34.77(9.14)
| 35.47(10.44)
| 33.94(0.08)
| 35.09(8.86)
| 0.977
| 0.942 |
Conclusion
In this study, there were no significant differences in times to peak, the mean peak intensity, and SD of pixel value in ROI (from the frame with the highest mean pixel value) among PDAC, PNET, and AIP, however there was significant difference in the mean peak intensity between PDAC and non-PDAC group. A study with larger test number is needed in the future for further application.
Disclosure
We have no conflicts of interest to disclose.
