Acceleration of Hyperspectral Skin Cancer Image Classification through Parallel Machine-Learning Methods.

Area of Science:

• Medical imaging
• Machine learning applications
• Computational science

Background:

• Hyperspectral imaging (HSI) is valuable in medicine for diagnostic support and surgery guidance.
• Processing hyperspectral data is computationally intensive, hindering rapid disease detection.
• There is a need for faster analysis of HSI data, especially in time-sensitive medical applications.

Purpose of the Study:

• To accelerate the classification of hyperspectral skin cancer images.
• To investigate the parallelization of Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGB) algorithms using Compute Unified Device Architecture (CUDA).
• To evaluate the performance of these parallelized algorithms on Graphical Processing Units (GPUs).

Main Methods:

• Parallelization of SVM, RF, and XGB machine learning algorithms using CUDA.
• Implementation of parallel algorithms on GPUs for hyperspectral image classification.
• Comparison of serial versus parallel algorithm execution times for hyperspectral skin cancer image analysis.

Main Results:

• Parallelized SVM and XGB algorithms demonstrated significant improvements in classification times compared to their serial versions.
• The study highlights the suitability of GPUs for accelerating hyperspectral data processing.
• All tested algorithms showed good performance in hyperspectral image classification, particularly with limited datasets.

Conclusions:

• Parallel computing using GPUs can effectively reduce the computational burden of hyperspectral image analysis.
• Accelerated classification of hyperspectral skin cancer images using parallel SVM and XGB is feasible and beneficial.
• This approach supports faster disease detection, addressing the demand for timely medical diagnoses.