The feasibility of simple Quantum Neural Network (QNN) architectures with multi-channel inputs for multi-class classification tasks is investigated by comparison of their performance to classic convolutional neural networks (CNNs) on general-purpose (CIFAR10) and medical (DermaMNIST/MedMNIST) datasets. The classical and quantum convolution operations are explored, utilizing the CIFAR10 and DermaMNIST datasets to assess the efficacy of these architectures. The QNN models integrate quantum operations, such as the Y-axis quantum transformation, to enhance the feature extraction process. Three models are evaluated: a baseline CNN (based on LCNet architecture as an example), a hybrid neural network (HNN) with four quantum convolutional channels (HNN-4QC), and a hybrid model with an additional input channel containing the original image (HNN-5QC). Our results demonstrate that HNN-4QC and HNN-5QC consistently outperform the baseline CNN model in terms of accuracy and area under the curve for receiver operating characteristics across both datasets, validating the potential of quantum-enhanced architectures for complex classification tasks. This study lays the groundwork for further exploration of quantum computing in neural network design for computer vision.