The integration of artificial intelligence (AI) in sectors such as technology and medical practice is very attractive, due to the development of a hybrid model including convolutional neural networks (CNN) and transformers. These models being optimized, their application is carried out for specific purposes such as 3D segmentation, pathology recognition, and technical diagnosis of information. In this article, we set up a comparative study that aims to analyze the effectiveness and limitations of these approaches, taking into account various parameters such as accuracy, computational performance and adaptability to complex environments. Studies show that these techniques allow these academics to significantly improve, both in accuracy and in computing efficiency, to oscillate in large information formats while reducing the power of calculations. However, and although these models have shown their value in their use cases, there remain many barriers to the adoption of these models on large scales. Among these challenges, it is about data management based on heterogeneous systems, closing the problem of using models on disparate environments, and difficulties of integration into existing industrial / clinical systems. Although many of these models have potential, extensive modifications are needed before these models are used in real-world situations.