In medical imaging, there are many challenges associated with diagnostic systems to detect tumors earlier: the integration of different imaging modalities. Due to the multifactorial nature of tumors, the lack of representation of features and the high proportion of false positives, the reliability of diagnosis and timely treatment may be compromised. Early detection of tumours is of special importance, as it is essential for increased patient survival, reduced health care costs, and timely clinical intervention. To overcome these difficulties, a Deep Pattern Recognition Framework (DPRF) for multimodal medical image analysis is introduced. It embeds multimodal feature fusion methods and uses them to efficiently fuse information from Magnetic Resonance Imaging, computed tomography, and positron emission tomography images within the framework. The approach employs multimodal feature fusion techniques to fuse features from Magnetic Resonance Imaging, computed tomography, and positron emission tomography (PET) images within the framework. Additionally, preprocessing, normalization, and attention-guided feature selection are added to further enhance feature quality and better discriminate tumors. The proposed framework was tested on a multimodal dataset of 12,500 patient scans to comprehensively evaluate across a range of different imaging conditions. Experimental results indicate that its accuracy, precision, recall, F1-score, specificity, and AUC are 98.64%, 98.21%, 98.47%, 98.34%, 98.76%, and 99.12%, respectively, significantly outperforming the conventional CNN, ResNet and Transformer-based approaches. The results indicate that DPRF can accurately detect tumors at an early stage and minimize misdiagnosis and false alarms. As such, the framework is reliable, scalable, and interpretable for intelligent medical imaging systems, aiding radiologists in quicker and more accurate clinical decisions.