Improvement of the Performance of Cell Counter Devices Using a Proposed Model Combining Multilayer Perceptron Neural Networks
N. Ahmadi, S. Asyabi · Transactions on Machine Intelligence · 2021
In automated cell counting devices, the performance of the counting channel can be significantly affected under problematic conditions, such as platelet aggregation, leading to inaccuracies in key blood parameter measurements. Given the limitations of existing algorithms in addressing these challenges, this study proposes an enhanced algorithm to improve the counting accuracy of critical blood components, particularly platelets and hematocrit, within the counting chamber. To achieve this, a hybrid approach integrating two computational models was implemented, demonstrating an improvement in overall counting performance. Among the tested optimization techniques, the Satin Bowerbird Optimization (SBO) Algorithm yielded superior results, outperforming other methods in terms of prediction accuracy. While the Biogeography-Based Optimization (BBO) Algorithm and the Teaching-Learning-Based Optimization (TLBO) Algorithm exhibited higher accuracy for certain blood parameters compared to the SBO Algorithm, the SBO Algorithm achieved the highest number of correct predictions across all parameters. In contrast, the Particle Swarm Optimization (PSO) and Firefly (FA) Algorithms failed to produce reliable results. The findings highlight the effectiveness of the proposed algorithm in enhancing the robustness and precision of blood parameter quantification, making it a promising approach for improving automated cell counting in clinical and laboratory applications.