Optifusion: advancing visual intelligence in medical imaging through optimized CNN-TQWT fusion

The domain of medical image fusion has garnered considerable attention within the biomedical imaging and clinical analysis communities. Despite the transformative impact of deep learning models on image fusion, achieving perfection remains an ongoing challenge. To address this challenge, a network named advancing visual intelligence in medical imaging through optimized CNN-TQWT fusion (OCT-fusion) is proposed to advance the field of multimodal medical image fusion by adeptly integrating spatial and transform domain strategies, ensuring superior image clarity. A modified CNN (MCNN) optimized for superior feature extraction from source images is proposed, and a tunable-Q wavelet transform (TQWT) is employed to refine the fusion process. This combination ensures that the MCNN-enhanced image is further integrated with a source image, reinforcing the overall clarity. The hyper-parameters of the TQWT undergo precision tuning using the covariance matrix adaptation evolution strategy (CMA-ES) optimizer. After applying the inverse TQWT (ITQWT), the final image captures the best from both domains, promising unparalleled clarity and detail. As a contribution in the realm of medical image fusion, the suggested approach when applied to medical and infrared (IV) datasets performed better than existing approaches which is evident in terms of seven performance measures. The novelty of the proposed method lies in the fact that it is the first one to apply TQWT decomposition in an image fusion task along with the optimization of TQWT parameters with Optuna. The method is tested on publicly available dataset, i.e., Whole Brain Atlas Harvard medical dataset (Johnson in The whole brain atlas, 2001). The robustness of the model is also checked on infrared–visual (IV) and color multi-focus datasets (Zhang et al. in Inf Fusion 54:99-118, 2020).