Kyan Matthew, Guan Ling, Liss Steven
School of Electrical and Information Systems Engineering, University of Sydney, NSW 2006, Australia.
Neural Netw. 2005 Jun-Jul;18(5-6):850-60. doi: 10.1016/j.neunet.2005.06.032.
The Self Organising Tree Map (SOTM) neural network is investigated as a means of segmenting micro-organisms from confocal microscope image data. Features describing pixel and regional intensities, phase congruency and spatial proximity are explored in terms of their impact on the segmentation of bacteria and other micro-organisms. The significance of individual features is investigated, and it is proposed that, within the context of micro-biological image segmentation, better object delineation can be achieved if certain features are more dominant in the initial stages of learning. In this way, other features are allowed to become more/less significant as learning progresses: as more knowledge is acquired about the data being segmented. We argue that the efficiency and flexibility of the SOTM in adapting to, and preserving the topology of input space, makes it an appropriate candidate for implementing this idea. We propose a refinement to the competitive search strategy that allows for a more appropriate fusion of signal and proximal features, thereby promoting a segmentation that is more sensitive to the regional associations of different microbial matter. A refined stop criterion is also suggested such that the dynamically generated number of classes becomes more data dependant. Preliminary experiments are presented and it is found that favouring intensity characteristics in the early phases of learning, whilst relaxing proximity constraints in later phases of learning, offers a general mechanism through which we can improve the segmentation of microbial constituents.
自组织树状图（SOTM）神经网络被作为一种从共聚焦显微镜图像数据中分割微生物的方法进行研究。研究了描述像素和区域强度、相位一致性和空间邻近性的特征对细菌和其他微生物分割的影响。研究了各个特征的重要性，并提出在微生物图像分割的背景下，如果某些特征在学习的初始阶段更占主导地位，就能实现更好的目标描绘。这样，随着学习的推进，其他特征会变得更重要或不那么重要：因为获得了更多关于被分割数据的知识。我们认为，SOTM在适应和保留输入空间拓扑结构方面的效率和灵活性，使其成为实施这一想法的合适候选者。我们对竞争搜索策略提出了一种改进，允许对信号和邻近特征进行更合适的融合，从而促进对不同微生物物质区域关联更敏感的分割。还提出了一个改进的停止标准，使动态生成的类别数量更依赖于数据。给出了初步实验结果，发现学习早期偏向强度特征，而学习后期放宽邻近性约束，提供了一种通用机制，通过该机制我们可以改进微生物成分的分割。
