AUTOMATION OF THE DIAPHRAGM TO REDUCE DOSE LOAD

Golya Roman Dmitrievich

Abstract: Objective: To reduce the radiation load while working on mobile X-ray surgical devices by automating the operation of the slit and iris diaphragm using a convolutional neural network. Methods: Application of the convolutional neural network UNet, trained on its own anonymized sets of X-rays and open datasets ChestX-ray and MIMIC-CXR. Results: During the training of the neural network, it was possible to achieve a value of 97% accuracy with an acceptable 33 milliseconds for frame processing, followed by a decision on the state of the aperture. Conclusions: The proposed method made it possible to get rid of the dependence in the operator when installing the diaphragm in the desired position and size, thereby achieving a more stable and accurate control over the dose load.

Keywords:CNN; deep learning; dose load; diaphragm; medicine; X-ray; segmentation of objects