Hong Yang Jiang1, He Ma, Wei Qian, Guo Hui Wei. 1. From the *Sino-Dutch Biomedical and Information Engineering School, Hunnan Campus, Northeastern University, Shenyang, China; and †College of Engineering, University of Texas at El Paso, El Paso, TX.
Abstract
OBJECTIVE: The purpose of this study is to design a content-based medical image retrieval system, which helps excavate and assess pathological change of pulmonary parenchyma for risks analysis. METHODS: A data set including lung computed tomography images obtained from 115 patients who experienced pathological changes in pulmonary parenchyma is used. Using morphological theory, images are preprocessed and decomposed into groups of pixel blocks (words), which construct vocabulary. A latent Dirichlet allocation (LDA) model is constructed to assess each image for risk analysis with the method of leave-one-out cross-validation. The precision and recall rate are used as the performance assessment criteria. RESULTS: The LDA model generates a relevance rank of retrieval results from high to low. From the top 50 images, precision of identical tissue is 0.76 ± 0.031 and precision of each attribute of pulmonary parenchyma range from 0.776 ± 0.043 to 0.984 ± 0.008. CONCLUSIONS: The study results demonstrate that the proposed LDA model is conductive to lung computed tomography image retrieval and has reliable efficacy on risk analysis about pathological changes of pulmonary parenchyma.
OBJECTIVE: The purpose of this study is to design a content-based medical image retrieval system, which helps excavate and assess pathological change of pulmonary parenchyma for risks analysis. METHODS: A data set including lung computed tomography images obtained from 115 patients who experienced pathological changes in pulmonary parenchyma is used. Using morphological theory, images are preprocessed and decomposed into groups of pixel blocks (words), which construct vocabulary. A latent Dirichlet allocation (LDA) model is constructed to assess each image for risk analysis with the method of leave-one-out cross-validation. The precision and recall rate are used as the performance assessment criteria. RESULTS: The LDA model generates a relevance rank of retrieval results from high to low. From the top 50 images, precision of identical tissue is 0.76 ± 0.031 and precision of each attribute of pulmonary parenchyma range from 0.776 ± 0.043 to 0.984 ± 0.008. CONCLUSIONS: The study results demonstrate that the proposed LDA model is conductive to lung computed tomography image retrieval and has reliable efficacy on risk analysis about pathological changes of pulmonary parenchyma.