Automated Medical Visualization: Application of Supervised Learning to Clinical Diagnosis, Disease and Therapy Management

Adekunle M. Adeshina, and M.A. Dikko

Keywords: Brain Tumour, Compute Unified Device Architecture (CUDA), Medical Visualization


The rapid advancement in high performance computing, ultrafast computing, autonomous technologies and complexity of biomedical data for visualization and image guidance play a significant role in modern surgery, redefining the ways in which surgeons perform their surgical procedures. Undoubtedly, several notable efforts have been recorded towards the development of a reliable computer-assisted medical visualization; unfortunately, aligning medical volume visualization applications with Artificial Intelligence features still remains challenging. Brain tumour diagnosis requires an enhanced, effective and accurate 3-D visualization system for navigations, references, diagnoses as well as documentations. With this study, a 3-D, high- performance artificial intelligence-enabled medical visualization framework was designed and implemented using automated machine learning features (AutoML), leveraging the auto-selective capabilities of machine learning to help specialists in the identification of the most appropriate regions of interest with their associated hyper parameters to optimizing performances, while simultaneously attempting to maximizing the reliability of the resulting predictions. C# and Compute Unified Device Architecture (CUDA) within the Microsoft.NET environment, in side-by-side comparison with VB.NET were used for the implementation. The framework was evaluated for rendering processing speed using the brain datasets obtained from the Department of Surgery, University of North Carolina, United States. Interestingly, the framework achieves 3-D visualization of the human brain, reliable enough to detect and locate possible brain tumour within high interactive speed and accuracy. Furthermore, in comparison with other previously proposed medical visualization systems, the framework was able to significantly highlight automatically, with greater accuracy, the features in brain datasets without any manual intervention.