Font Size: a A A

Modeling Temporal Information Of Mitosis For Mitotic Event Detection

Posted on:2019-06-15Degree:MasterType:Thesis
Country:ChinaCandidate:H Y ChengFull Text:PDF
GTID:2370330593451650Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Due to the importance of the study of cell division and the advances of machine learning in behavior and event detection,many scholars hope to use machine learning to study the life activities of cells in vitro.At present,the mitosis recognition and cell cycle location methods can be divided into three categories: trajectory based methods,feature based methods and temporal information modeling based methods.In these methods,the trajectory methods is too hard to achieve;feature based methods cannot capture dynamic information in events effectively;for the above two methods,the third type methods perform better,but needs a large number of training samples to learn,which leads to low efficiency.From the summary of the previous work,and the exist problems of mitosis recognition and cell cycle location,in view of these problem,the paper put forwards a method based on ranking,which based on image sequence with extending the mitotic event to the video sequence event,and taking the ordered information as the representation of the video sequence event.Firstly,based on the visual feature,a vector function is constructed by pooling operation to simulate the data distribution of image sequence,which can capture the sequence change direction in continuous time,and make the visual feature and the time variable interrelated,this time meaning order can help a lot for rank learning.Secondly,we train a linear ranking machine to integrate the historical information in the image sequence,and use the parameter of the rank function to represent the sequence,as the new feature is sequence specific which can effectively capture the dynamic evolution of the sequence.Finally,a binary SVM classifier is trained on the new feature to complete the sequence classification.The proposed method is simple and efficient with a sequence can be represent by a feature vector.We capture the phase characteristic of the sequence by dividing the sequence into subsequences and extracting the new feature based on these subsequences.The sequence stage labels obtained by training a SVM multi classifier to achieve cell cycle location.In this paper,the mitotic events recognition and localization experiments were performed on the most chalenging C3H10T1/2 and C2C12 datasets in the field of cell event detection.The experimental results show that the proposed sequence features can effectively simulate the data distribution of image sequences and depict the phase characteristics of mitotic events.The experimental results show that the proposed method out performed than the state of the art method.
Keywords/Search Tags:Mitosis recognition, Mitosis detection, Sequential image analysis
PDF Full Text Request
Related items