Sensors & Transducers

Vol. 249, Issue 2, February 2021, pp. 1-8

1 Markus GERLOFSMA, 1, 2, * Milan PETKOVIC, 2 Peter van LIESDONK
and 1 Vlado MENKOVSKI

1 Eindhoven University of Technology, Eindhoven, The Netherlands

2 Royal Philips Research Laboratories, Eindhoven, The Netherlands

* E-mail:

Received: 10 October 2020 /Accepted: 10 December 2020 /Published: 28 February 2021

Abstract: The acquisition of Computed Tomography (CT) images has shown to be affected by the scanning device which acquires the image. Specifically, the same subject, scanned by a variety of CT devices, holds different properties depending on the device which acquired the image. This variance consequently has an impact on the medical analysis of the images. Therefore, the ability to determine the acquiring CT device based on the produced CT images may lead to improved diagnoses. In addition, knowledge of the CT device may furthermore provide a means to ensure verification of provenance without the necessity to rely on potentially corrupt or missing DICOM metadata. In this work, we apply a Convolutional Neural Network (CNN) to classify 4 manufacturers of CT scanners, based on the CT images which their devices generate. We apply our experiments on a large, publicly available dataset, and additionally apply previous classification techniques on the same dataset. With an accuracy of up to 93.6 %, our approach significantly outperforms existing work.

Keywords: CT scanners, CT Device Manufacturers, CT Image Classification, Convolutional Neural Networks, Sensor Noise.