RGB Image analysis of pancreatic mass-elastographies in endoscopic ultrasound (EUS) can predict malignancy- a pilot study

INTRODUCTION Image analysis using RGB (red-green-blue) image profiling is a well-accepted technique, especially in histopathological image analysis.

AIM To investigate the accuracy of quantitative image analysis of EUS-elastographies (tissue elasticity) to predict malignancy of solid pancreatic lesions.

METHODS Elastographies of solid pancreatic masses obtained with EUS between 01/2014-06/2019 were extracted from our ultrasound device (Hitachi V70) and analyzed retrospectively. Quantitative RGB based analysis was performed using a Java image processing program (ImageJ, NIH). Red indicates soft tissue, blue hard tissue, and green intermediate tissue-elasticity. The exact amount of blue, red and green color was measured and expressed in pixels and percentages. After calibration of each image, the color intensity was measured on a scale of 0-255 for an 8-bit image. The intensity ratio for each color was defined as the relation between absolute value for this color and the intensity of the sum of all three colors (R+G+B). The tissue surrounding the tumor outside the well-defined margins of the lesion on ultrasound was not included in the analysis. The final diagnosis was made either by histopathology or radiological findings in combination with tumor markers and clinical follow-up.

RESULTS 59 solid pancreas tumors evaluated by strain elastography were analyzed: 45 (75 %) malignant (60 % adenocarcinoma, 8.3 % metastasis and 6.6 % neuroendocrine tumors) and 14 (23.3 %) benign masses. Cut-offs values to differentiate between malignant and benign pancreatic tumors were calculated for parameters with good correlation for the presence of malignancy (criteria, Table). Risk of malignancy according to these criteria was: 4/4 (15 cases)-100 %, 3/4 (24 cases)- 87.5 %, combined 3 or 4/4 (39 cases)-92.3 %; 2/4(11 cases)-54.5 %, 1/4(4 cases)-75 % and 0/4 (5 images)-0 %.

CONCLUSION Quantitative image analysis of solid pancreatic lesion elastographies obtained in EUS may predict (3 or 4/4 criteria) or exclude (0/4 criteria) malignancy with high accuracy.

Tab. 1

Parameter (criteria)	Cut-offs (%)	Sensitivity (%)	Specificity (%)	AUC
Blue color (%)(Spearman r = 0.192,p = 0.10)	>  55	93.3	35.7	0.62
Green color (%)(r=-0.218,p = 0.09)	<  42.5	97.8	42.5	0.64
Green color Intensity Ratio(r=-0.391,p = 0.002)	<  56	71.1	78.6	0.76
Red color Intensity Ratio(r=-0.194,p = 0.10)	<  18.5	42.2	92.9	0.63

Publikationsverlauf

Artikel online veröffentlicht:
26. Mai 2020