Skip to Main Content

An official website of the United States government

Development and validation of a clinical decision support system for diagnosis of solitary pulmonary nodules using image-based features and clinical information

Principal Investigator
Name
Yuechen Qian
Degrees
Ph.D.
Institution
Philips Research North America
Position Title
Principal Scientist
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-52
Initial CDAS Request Approval
Feb 26, 2014
Title
Development and validation of a clinical decision support system for diagnosis of solitary pulmonary nodules using image-based features and clinical information
Summary
NLST showed a benefit with significant mortality reduction with low-dose CT (LDCT) screening of high-risk persons. However, 95 % of all positive findings do not lead to the diagnosis of lung cancer. Thus, clinical decision support tools, such as computer-aided detection/diagnosis and other tools could aid the physicians to resolve false-positive results and decrease patient anxiety. We aim to develop and validate a clinical decision support system, which will include a computer-aided lesion detection using LDCT, estimation the likelihood of malignancy of a detected solitary pulmonary nodule (SPN), presentation of similar cases from a database of known-diagnosis, automatic tracking of lesion for follow-up, and reliable estimate of nodule characteristics.
Aims

1. Validate the performance and further refine our proprietary computer-aided detection algorithm using the NLST CT data.
2. Further develop a computer-aided diagnosis algorithm to predict the malignancy of a detected SPN. Image-based features, such as location, size, shape, margin, fissure involvement, texture, etc. will be calculated from the LDCT of the SPNs. After fusing these image-based features with clinical information (age, sex, family history, emphysema, etc.) of the patient, machine learning techniques, such as committee of classifiers will be trained to provide a malignancy likelihood of a SPN.
3. Develop an algorithm to retrieve similar nodules from a database of patients with known diagnosis in terms of pathology and perceptual similarity. The retrieval database will be formed from a subset of patients having malignant and benign nodules in the NLST data and the corresponding diagnosis based on pathology result or 2-year stability of the finding. The nodules/patients will be characterized by image-based features (size, shape, texture, etc.) and clinical information (age, sex, family history, emphysema, etc.). Machine learning techniques (e.g. feature selection and distance calculation) will be applied to retrieve the most similar cases.
4. Validate the performance of the above developed algorithms (classification accuracy, sensitivity, specificity, PPV, NPV) on the subset of the NLST data, which was not used for the development of the algorithms.
5. Develop and validate algorithms for automated detection and correlation of nodules in follow-up studies for high-precision growth rate estimation.
6. Estimate image-based characteristics, such as location, type, size, shape, margin, fissure involvement, etc. of SPNs using image processing and machine learning techniques and validate the performance of the estimation with respect to physicians’ annotations in the NLST data.

Collaborators

Michael C. Lee, Senior Member Research Staff, Philips Research, Briarcliff Manor, USA
Rafael Wiemker, Senior Scientist, Philips Research, Hamburg, Germany
Tobias Klinder, Senior Scientist, Philips Research, Hamburg, Germany
Ekta Dharaiya, Research Scientist, Philips Healthcare, Cleveland, USA
Amnon Steinberg, Clinical Solutions specialist, Philips Healthcare, Haifa, Israel