Cancer Genetics and Epigenetics 2024, Vol.12, No.4, 210-222 http://medscipublisher.com/index.php/cge 215 by combining clinical and radiomics features, demonstrating the effectiveness of feature selection in enhancing model accuracy (Li et al., 2019). 4.3.2 Data normalization and standardization Data normalization and standardization are critical preprocessing steps that ensure the comparability of data from different sources. These techniques adjust the scale of data, making it suitable for machine learning algorithms. For instance, a study on brain tumor segmentation prediction in MRI scans employed feature fusion and joint label fusion methods, which involved normalizing multi-resolution texture features and tumor cell density features to improve segmentation accuracy (Pei et al., 2020). 4.4 Tools and platforms for AI-Based data fusion Several tools and platforms have been developed to facilitate AI-based data fusion in medical research. These platforms provide the necessary infrastructure for data integration, model training, and validation. TensorFlow and PyTorch: These are popular open-source deep learning frameworks that support the development and deployment of AI models. They offer extensive libraries for building and training neural networks, making them suitable for multi-modal data fusion tasks. Prediction One: An auto-AI software used to predict colon cancer recurrence with improved accuracy compared to conventional statistical models. This tool allows clinical surgeons to construct AI models without extensive knowledge of AI (Mazaki et al., 2021). The Cancer Genome Atlas (TCGA): A comprehensive database that provides multi-modal cancer data, including genomic, clinical, and imaging data. Researchers can leverage TCGA data to develop and validate AI models for cancer prediction and prognosis (Shao et al., 2020; Vale-Silva and Rohr, 2021). In conclusion, the integration of multi-modal data using AI techniques holds great promise for improving colon cancer prediction. By leveraging deep learning models such as CNNs and RNNs, along with statistical and computational methods for feature selection and data normalization, researchers can develop accurate and effective prediction models. The availability of advanced tools and platforms further facilitates the implementation of AI-based data fusion in clinical practice. 5 Applications of AI in Multi-Modal Data Fusion for Colon Cancer Prediction 5.1 Predictive models and algorithms 5.1.1 Supervised learning models Supervised learning models have been extensively used in the prediction of colon cancer by leveraging multi-modal data. For instance, a study demonstrated the use of a multi-scale feature fusion convolutional neural network (MFF-CNN) based on shearlet transform to identify histopathological images of colon cancer. This model achieved an identification accuracy of 96% and an average F-1 score of 0.9594, significantly reducing false negative and false positive rates (Liang et al., 2020). Another example is the development of a novel prediction model for colon cancer recurrence using auto-artificial intelligence (AI), which showed improved accuracy compared to conventional statistical models, with an AUC of 0.815 (Mazaki et al., 2021). 5.1.2 Unsupervised learning models Unsupervised learning models, although less common in direct prediction tasks, play a crucial role in feature extraction and data preprocessing. These models can identify patterns and structures in multi-modal data without labeled outcomes, which can then be used to enhance the performance of supervised learning models. For example, clustering techniques can be used to group similar histopathological images or genomic data, which can then be fed into supervised models for improved prediction accuracy (Thakur et al., 2020). 5.1.3 Semi-Supervised and reinforcement learning models Semi-supervised and reinforcement learning models combine the strengths of both supervised and unsupervised learning. These models are particularly useful when labeled data is scarce. A multi-task multi-modal feature
RkJQdWJsaXNoZXIy MjQ4ODYzNQ==