This manuscript presents the Quantum-inspired Complex Convolutional Neural Networks (QICCNN) method for COVID-19 diagnosis using lung CT scan images. The QICCNN method utilizes three transfer learning techniques: ResNeXt-50, CoAtNet, and EfficientNetV2 for COVID-19 diagnosis. Experiments are conducted under three strategies to evaluate the performance of ResNeXt-50, CoAtNet, and EfficientNetV2 using optimization algorithms such as Binary Chimp Optimization Algorithm (BChOA), Tyrannosaurus Optimization Algorithm (TROA) and Dipper Throated Optimization Algorithm (DTOA). Three strategies are employed for training and testing the QICCNN methods. In strategy 1, QICCNN methods are trained and tested using a single-source dataset. In strategy 2, QICCNN methods are trained using a single-source dataset and tested using a multisource dataset. In strategy 3, QICCNN methods are trained and tested using a multisource dataset. Single-source datasets are acquired from a single geographical location, and multisource datasets are collected from various geographical locations. The CT scan images are pre-processed with Distributed Set-Membership Fusion Filtering (DSMFF), and then features are extracted using Two-Sided Offset Quaternion Linear Canonical Transform (TSOQLCT). Results indicate that all three QICCNN models performed better with the BChOA optimizer than with TROA and DTOA. The ResNeXt-50 model with BChOA outperformed CoAtNet-BChOA and EfficientNetV2-BChOA in COVID diagnosis.The performance of the proposed QICCNN-ResNeXt-50-BChOA-CSM approach shows higher accuracy, precision, sensitivity, F1-score, and ROC compared to existing methods such as LCTS-SqueezeNet-CSM, CTLI-VGG16-CSM, and CTAS-DarkNet19-CSM models respectively. The analysis indicates that the proposed QICCNN-ResNeXt-50-BChOA-CSM method can be integrated with CT scan devices and can save precious time for radiologists in diagnosing COVID diseases. © 2024 Elsevier B.V., All rights reserved.