With the popularity of social media and online platforms, fake news has spread faster and more widely than ever before, with far-reaching impacts on many fields such as politics, economics, and society. Therefore, the research and development of fake news detection technology is particularly important. At present, the research in the field of fake news detection mainly focuses on the fusion and consistency detection of images and texts, and remarkable results have been achieved in this direction. However, the existing methods still face two major challenges: first, how to integrate multimodal features more evenly to ensure that the image and text are equal in the model; The second is how to construct a fusion model that can effectively deal with the potential inconsistencies between images and texts. In order to solve the above challenges, this paper proposes a Collaborative Attention Network Based on Multimodal Inconsistency (MICAN) for fake news detection. MICAN can not only fuse the features of text and images, but also effectively solve the problem of image-text mismatch that may occur during the fusion process. Through the innovative multi-modal collaborative attention network framework and mutual attention neural network architecture, the network realizes the accurate fusion of multimodal features, and enhances the accuracy and robustness of fake news detection. In addition, the MICAN model also adds an event discriminator, which is responsible for predicting the labels of events to achieve accurate classification and identification of news events, which further improves the practicability of the model. Experimental results show that the MICAN model outperforms the existing fake news detection models on multiple datasets such as weibo, twitter, and PHEME, which verifies its effectiveness and superiority.