Abstract

Background: The diagnosis of diseases correctly became a challenge, and any error can cost patients life, especially when there is a lack of knowledge or expertise related to a disease, it often results in patient’s death or takes the form of an epidemic, as we have seen in the case of Ebola.

Objective: The automation and the development of reliable diagnostic systems became a necessity. Through the use of technology, we can automatically share the knowledge without formal interaction as well as we can identify areas where the disease is spreading while it is not known by the doctors there.

Methods: We have presented a complete system that utilizes a combination of one of the best techniques in the field of parallelism, classification, and knowledge sharing. We have used two data sets (DDSM and Belarus Tuberculosis data) to test the applicability of the idea. After retrieving the data, the images are preprocessed, and then Gray level co-occurrence matrix features have been extracted and finally passed to training using three versions of support vector machines.

Results: GPU-Accelerated SVM outperformed both parallelized SVM and sequential SVM using breast cancer data, but with lung CT images, GPU-accelerated LIBSVM have not given a remarkable speed-up because the data is small and the gain is lost due to the gpu-cpu memory and cpu-gpu transfer time. The accuracy performances given by three SVMs were identical.

Conclusion: Automation through knowledge sharing and parallel computing can help to deal across the world with diseases and it will be easy for doctors to draw the inference.

Keywords: Diseases identification, image classification, image processing, knowledge sharing, parallel support vector machines, parallelism.

Graphical Abstract