Enhanced Moth-flame Optimizer with Quasi-Reflection and Refraction Learning with Application to Image Segmentation and Medical Diagnosis

Abstract

Background: Moth-flame optimization will meet the premature and stagnation phenomenon when encountering difficult optimization tasks. Objective: To overcome the above shortcomings, this paper presented a quasi-reflection moth-flame optimization algorithm with refraction learning called QRMFO to strengthen the property of ordinary MFO and apply it in various application fields. Method: In the proposed QRMFO, quasi-reflection-based learning increases the diversity of the population and expands the search space on the iteration jump phase; refraction learning improves the accuracy of the potential optimal solution. Results: Several experiments are conducted to evaluate the superiority of the proposed QRMFO in the paper; first of all, the CEC2017 benchmark suite is utilized to estimate the capability of QRMFO when dealing with the standard test sets compared with the state-of-the-art algorithms; afterward, QRMFO is adopted to deal with multilevel thresholding image segmentation problems and real medical diagnosis case. Conclusion: Simulation results and discussions show that the proposed optimizer is superior to the basic MFO and other advanced methods in terms of convergence rate and solution accuracy.

Motion-artifact-augmented pseudo-label network for semi-supervised brain tumor segmentation

“…In the past few years, the medical image segmentation method based on machine learning and deep learning remarkably improved the image segmentation accuracy (Zhao et al 2018, Xu et al 2021, Zhang et al 2021, Hu et al 2022, Ren et al 2022, Liu et al 2023a, Ye et al 2023. However, deep learning models demand substantial amounts of data, necessitating larger and higher-quality labeled datasets for accuracy improvement.…”

YOSEG Enhancing Instance Segmentation of Medical Small Targets

“…Accurate medical image segmentation is pivotal in various clinical applications. The continuous progress in deep learning has yielded numerous algorithms capable of achieving precise medical image segmentation, thereby facilitating the process [1][2][3][4][5][6][7][8][9]. However, medical small objects pose significant challenges to deep learning-based segmentation due to their intricate structures, limited contrast, and potential overlap with surrounding tissues [10].…”