研究目的
To propose an image segmentation method based on a modified simplified pulse-coupled neural network (MSPCNN) that incorporates human visual system (HVS) characteristics to improve segmentation precision and reduce computational complexity for medical images.
研究成果
The proposed MSPCNN method with HVS-based stimulus input achieves higher segmentation accuracy and lower computational complexity compared to state-of-the-art methods, as evidenced by OEM values of 0.8784 for MIAS, 0.8606 for DDSM, and 0.8585 for gallstones. Future work will explore applications in image qualification and enhancement.
研究不足
The method may have limitations in handling very complex or noisy medical images, and the parameter settings rely on Otsu thresholding, which might not be optimal for all image types. Computational efficiency could be further improved.
1：Experimental Design and Method Selection:
The study uses a modified simplified pulse-coupled neural network (MSPCNN) model with parameters set automatically based on Otsu thresholding. A neighboring firing matrix Q is introduced to determine stimulus inputs based on HVS principles.
2：Sample Selection and Data Sources:
Medical images from the Mammographic Image Analysis Society (MIAS) database, Digital Database for Screening Mammography (DDSM), and ultrasound images of gallstones from Gansu Provincial Hospital are used.
3：List of Experimental Equipment and Materials:
MATLAB R2014a software on a computer with Intel(R) Core(TM) i7-7700HQ processor and 8G DDR3 RAM.
4：Experimental Procedures and Operational Workflow:
Images are processed by calculating normalized Otsu thresholding, computing neighboring firing matrix Q, determining sub-intensity range parameter Sint, deriving stimulus input Sioij using Weber-Fechner law, and applying MSPCNN for segmentation.
5：Data Analysis Methods:
Segmentation accuracy is evaluated using metrics such as uniformity (UM), mean intensity, standard deviation (STD), area overlap (OV), sensitivity (SEN), and an overall evaluation metric (OEM).
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
