Fig. 3
From: Deep learning for semi-automated unidirectional measurement of lung tumor size in CT
Overview of deep learning algorithm to perform unidirectional lesion measurement. In the shown example, the input image was augmented into 16 images at various magnifications, using the arbitrary input point (stained in red, pointed by arrow) in the target lesion. The neural network classified each augmented image whether the containing lesion size is larger or smaller than 32 pixels within a 128-by-128 pixel frame. Upon identification of a magnification inducing classification failure, the magnification and pixel spacing information were used to determine the final measurement