Quantum convolutional neural networks for malaria cell classification: A comparative study with classical CNNs

¹ Biophotonics, Council for Scientific and Industrial Research, South Africa
² Department of Human Biology, University of Cape Town, South Africa
³ College of Graduate Studies, University of South Africa, South Africa

^* Corresponding author: kmpofu@csir.co.za

Abstract

This work outlines a quantum convolutional neural network (QCNN) model designed to accurately detect and classify biomedical images. In this work, a classical CNN and QCNN algorithm were developed and compared in diagnosing malaria infected cells from noninfected cells. The QCNN performance metrics in this work were compared with the performance of the classical CNN algorithm. QCNN algorithms can potentially overcome some limitations of their classical counterpart algorithm, i.e. CNN. The theoretical computational complexity of a single convolutional layer in a CNN is O (N × k²), where N is the number of input data points, and k is the size of the convolutional kernel, whereas the theoretical computational complexities as low as O(log(N)) for certain operations, leveraging quantum parallelism to process high-dimensional data more effectively. In this work the authors compared the performance of the CNN and QCNN for a small malaria dataset. The preliminary results of this work show that CNN outperforms QCNN in terms of accuracy, the CNN had a peak accuracy of 75% whereas the QCNN had an accuracy of 54%. The finding of this work can have an impact on quantum computing and quantum machine learning techniques in medical imaging. In this work we find that QCNNs in their current state of development do not outperform CNNs.