Speaker
Description
Recently, Convolutional Neural Networks (CNNs) [1] have achieved state-of-the-art performance in many areas including medical sciences, and are the method of choice commonly used for data recognition or classification. CNNs have proven to work most efficiently on 2-dimensional data that are in form of images.
In case of Positron Emission Tomography (PET) [2,3] studies, CNN may be applied directly to the reconstructed distribution of radioactive tracer injected to the patient's body, as for example a pattern recognition tool. Nonetheless, much PET data still exists in non-image format and therefore
opens challenging research questions on whether they can be effectively trained using CNN. Examples of such tasks are estimation of time-of-flight from signals registered in scintillators [4] or classification of coincidence events acquired by PET scanner [5].
The goal of this presentation is the introduction of scheme of non-image data transformation into 2-dimensional matrices, as a preparation stage for classification based on CNNs. The first work to apply CNN on different kinds of non-image datasets, e.g., gene expression or text information, was
proposed in [6]. Here, we will focus mainly on the problem of processing of vectors with small number of features in comparison to the number of pixels in the output images. As an example, a discussion of application of the proposed methodology to classification of PET coincidence events will provided [7].
References
[1] Lecun Y, Bengio Y and Hinton G. Deep learning, Nature vol. 521, pp. 436, 2015.
[2] Humm J L, Rosenfeld A, Del Guerra A. From PET detectors to PET scanners,
European J. of Nucl. Med. & Mol. Imag. vol. 30, pp. 1574, 2003.
[3] Bailey D L. Positron Emission Tomography: Basic Sciences, Springer-Verlag, New York, 2005.
[4] Berg E and Cherry S R. Using convolutional neural networks to estimate time-of-flight from PET detector waveforms, Phys. Med. Biol., vol. 63, pp. 02LT01, 2018.
[5] Bielecki J. Application of the machine learning methods to the multi-photon event classification in the J-PET scanner, Msc thesis, Warsaw University of Technology, 2019.
[6] Sharma A, Vans E, Shigemizu D, Boroevich K A and Tsunoda T. Deepinsight: A methodology transform a non-image data to an image for convolution neural network architecture. Scientific Reports vol. 9, pp. 11399, 2019.
[7] Konieczka P. Convolutional Neural Networks in classification of multi-photon coincidences in J-PET scanner, 1st Symposium on Theranostics, 9-11 October 2021.
