Real time decision support system for diagnosis of rare cancers, trained in parallel, on a graphics processing unit.

In the present study a new strategy is introduced for designing and developing of an efficient dynamic Decision Support System (DSS) for supporting rare cancers decision making. The proposed DSS operates on a Graphics Processing Unit (GPU) and it is capable of adjusting its design in real time based on user-defined clinical questions in contrast to standard CPU implementations that are limited by processing and memory constrains. The core of the proposed DSS was a Probabilistic Neural Network classifier and was evaluated on 140 rare brain cancer cases, regarding its ability to predict tumors' malignancy, using a panel of 20 morphological and textural features Generalization was estimated using an external 10-fold cross-validation. The proposed GPU-based DSS achieved significantly higher training speed, outperforming the CPU-based system by a factor that ranged from 267 to 288 times. System design was optimized using a combination of 4 textural and morphological features with 78.6% overall accuracy, whereas system generalization was 73.8%±3.2%. By exploiting the inherently parallel architecture of a consumer level GPU, the proposed approach enables real time, optimal design of a DSS for any user-defined clinical question for improving diagnostic assessments, prognostic relevance and concordance rates for rare cancers in clinical practice.