A Master’s thesis entitled “Skin Cancer Diagnosis Using Pre-Trained Hybrid Convolutional Neural Network Models” was explored at the College of Computer Science and Information Technology, Al-Qadisiyah University. The thesis was presented by researcher Zahraa Saad Abdul Wahid Hassan, under the supervision of Assistant Professor Dr. Zahraa Jafat Alawi and Dr. Rasha Falah Kadhim.

The thesis aimed to present a hybrid model for skin cancer diagnosis and classification. This was achieved by utilizing deep learning, transfer learning, and feature fusion techniques, along with improving decision-making using fuzzy logic. This approach helps address the visual similarity between skin lesions and the significant variation within a single category, thereby enhancing the reliability of the classification process.

The experimental results showed that the three-part hybrid model combining DenseNet201, EfficientNetV2B2, and ResNet50 achieved the best performance. The classification accuracy reached 90.61% before decision optimization using fuzzy logic, and the accuracy of the proposed framework increased to 91% after decision optimization. The first frame (92%) and the second frame (92%) both achieved a Macro F1-Score, confirming the effectiveness of deep feature integration and decision optimization using fuzzy logic.

The thesis proposes several enhancements and smart techniques, which are expected to be future research projects with better results and higher performance. These include using larger and more diverse dermatology databases from different clinical institutions, testing the model in various clinical settings, exploring advanced deep learning architectures such as Attention-Based Networks, Vision Transformers, and Xception, and improving feature integration and assessment techniques in broader clinical environments.