On Cell Detection System with User Interaction

Cell detection is the essential step for various analyses in biological fields. One of the conventional approach is to construct a specialized method for every specified task, but it is not efficient in the meaning of the coding workload. Another approach is based on some machine learning technique, but it is difficult to prepare many training data. To solve these problems, we propose a balanced system by combining image processing and machine learning. The system is universally applicable to any image, because it only consists of basic methods of image processing. The code of the system does not need to be modified, because its behavior is adaptively tuned by machine learning. Users are free from excessive request of training data, because only a few desirable data is specifically requested by the system. The system consists of three units to achieve functionalities for avoiding parameter collision, compensating lack of training data, and reducing complexity of feature space. The effectiveness of the system is evaluated with a typical set of cell images, and the result is sufficient. The proposed system provides a reasonable way of preparing a tuned cell detection method for arbitrary sets of images in this field.