Design And Improvement Of Pulmonary Nodular Image Retrieval Algorithm Based On Learning To Rank

Doctors often use proven medical images when diagnosing new cases with medical images.Pulmonary nodule image retrieval is designed to retrieve candidate images that are similar to the new case image accurately for the doctor,and the purpose of the learning to rank algorithm is to make a more accurate sort of the search results.In order to establish an accurate and effective model of pulmonary nodule image retrieval,the features extracted from pulmonary nodule images need to be selected.After the feature subset is selected,the LambdaMART algorithm is used for the pulmonary nodule image retrieval task.The learning to rank algorithm is combined with the traditional content-based medical image retrieval algorithm.According to the features and the similarity marked by the professional doctors,the LambdaMART model is established to optimize the relative positional relationship between candidate images and NDCG evaluation criteria.In the prediction stage,the training model is used to measure the similarity between the new query image and the candidate image,and the candidate image with high similarity is selected.As single decision tree is hard to control the variance in the process of each round of iteration,The random forest algorithm is combined with the LambdaMART algorithm,and the single forest decision tree in the original LambdaMART model is replaced by the random forest model.On the other hand,the learning rate for the LambdaMART algorithm in a round of iterative process is fixed,and the learning rate is hard choose.Variable learning rate method is used to replace the original LambdaMART algorithm's fixed learning rate in Adaptive LambdaMART algorithm for adjustment learning rate with iteration number.The LambdaMART algorithm,the LambdaMARF and the Adaptive LambdaMART algorithm are used to analyze the pulmonary nodule image retrieval under a variety of feature selection algorithms.The experimental results show that the LambdaMART algorithm can good deal with the pulmonary node image retrieval task.The LambdaMARF algorithm and the Adaptive LambdaMART algorithm can further improve the accuracy of pulmonary nodal image retrieval.