| Furkan Atban, Muhammed Yusuf, Cüneyt Bayılmış Quantum Transfer Learning with Pretrained CNN Features for Multi-Class Medical Image Classification |
|---|
| Abstract. The automated classification of Optical Coherence Tomography (OCT) images is critical for the early diagnosis of retinal diseases such as choroidal neovascularization, diabetic macular edema and drusen. Classical deep networks achieve high accuracy but often require millions of parameters and significant computational resources. We present a hybrid Quantum Transfer Learning (QTL) framework combining a fine-tuned ResNet34 encoder with a variational quantum circuit (VQC) classifier for multi-class OCT classification. The classical backbone is partially fine-tuned—its first two layers remain frozen while the third and fourth blocks are adapted to OCT textures—and a classical projection reduces the 512-dimensional feature vector to a 5-dimensional input for a VQC acting on five qubits. Hyperparameters such as qubit count and circuit depth are selected via Particle Swarm Optimization (PSO). Experiments on the OCT2017 dataset show that the proposed model achieves 89.38% accuracy, supporting the feasibility of a parameter-efficient hybrid alternative for medical image classification under NISQ-inspired constraints. The results support the feasibility of hybrid quantum–classical models for medical imaging in the Noisy Intermediate-Scale Quantum era. |
| Keywords: Quantum Transfer Learning, Medical Image Classification, Optical Coherence Tomography, Hybrid Neural Networks, Particle Swarm Optimization |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2026.2.02 |