An image processing device includes at least one processor having hardware. The processor is configured to implement analyzing, in an intraductal image which is an image of inside of a duct, an intraductal imaging situation based on a position of a subject relative to a part that captures an image of the subject. Next, the processor is calculating a plurality of specific region discrimination indices for the intraductal image. Finally, the processor is detecting a specific region by integral determination on the plurality of specific region discrimination indices depending on the intraductal imaging situation.