Domain knowledge acquired from pilot studies is important for medical diagnosis. This paper leverages the population-level domain knowledge based on the D-optimal design criterion to judiciously select CT slices that are meaningful for the disease diagnosis task. As an illustrative example, the diagnosis of idiopathic pulmonary fibrosis (IPF) among interstitial lung disease (ILD) patients is used for this work. IPF diagnosis is complicated and is subject to inter-observer variability. We aim to construct a time/memory-efficient IPF diagnosis model using high resolution computed tomography (HRCT) with domain knowledge-assisted data dimension reduction methods. Four two-dimensional convolutional neural network (2D-CNN) architectures (MobileNet, VGG16, ResNet, and DenseNet) are implemented for an automatic diagnosis of IPF among ILD patients. Axial lung CT images are acquired from five multi-center clinical trials, which sum up to 330 IPF patients and 650 non-IPF ILD patients. Model performance is evaluated using five-fold cross-validation. Depending on the model setup, MobileNet achieved satisfactory results with overall sensitivity, specificity, and accuracy greater than 90%. Further evaluation of independent datasets is underway. Based on our knowledge, this is the first work that (1) uses population-level domain knowledge with optimal design criterion in selecting CT slices and (2) focuses on patient-level IPF diagnosis.

Lack of classifier robustness is a barrier to widespread adoption of computer-aided diagnosis systems for computed tomography (CT). We propose a novel Robustness-Driven Feature Selection (RDFS) algorithm that preferentially selects features robust to variations in CT technical factors. We evaluated RDFS in CT classification of fibrotic interstitial lung disease using 3D texture features. CTs were collected for 99 adult subjects separated into three datasets: training, multi-reconstruction, testing. Two thoracic radiologists provided cubic volumes of interest corresponding to six classes: pulmonary fibrosis, ground-glass opacity, honeycombing, normal lung parenchyma, airway, vessel. The multi-reconstruction dataset consisted of CT raw sinogram data reconstructed by systematically varying slice thickness, reconstruction kernel, and tube current (using a synthetic reduced-tube-current algorithm). Two support vector machine classifiers were created, one using RDFS ("with-RDFS") and one not ("without-RDFS"). Classifier robustness was compared on the multi-reconstruction dataset, using Cohen's kappa to assess classification agreement against a reference reconstruction. Classifier performance was compared on the testing dataset using the extended g-mean (EGM) measure. With-RDFS exhibited superior robustness (kappa 0.899-0.989) compared to without-RDFS (kappa 0.827-0.968). Both classifiers demonstrated similar performance on the testing dataset (EGM 0.778 for with-RDFS; 0.785 for without-RDFS), indicating that RDFS does not compromise classifier performance when discarding nonrobust features. RDFS is highly effective at improving classifier robustness against slice thickness, reconstruction kernel, and tube current without sacrificing performance, a result that has implications for multicenter clinical trials that rely on accurate and reproducible quantitative analysis of CT images collected under varied conditions across multiple sites, scanners, and timepoints.

Quantitative image analysis (QIA) goes beyond subjective visual assessment to provide computer measurements of the image content, typically following image segmentation to identify anatomical regions of interest (ROIs). Commercially available picture archiving and communication systems focus on storage of image data. They are not well suited to efficient storage and mining of new types of quantitative data. In this paper, we present a system that integrates image segmentation, quantitation, and characterization with database and data mining facilities. The paper includes generic process and data models for QIA in medicine and describes their practical use. The data model is based upon the Digital Imaging and Communications in Medicine (DICOM) data hierarchy, which is augmented with tables to store segmentation results (ROIs) and quantitative data from multiple experiments. Data mining for statistical analysis of the quantitative data is described along with example queries. The database is implemented in PostgreSQL on a UNIX server. Database requirements and capabilities are illustrated through two quantitative imaging experiments related to lung cancer screening and assessment of emphysema lung disease. The system can manage the large amounts of quantitative data necessary for research, development, and deployment of computer-aided diagnosis tools.

Prefetching methods have traditionally been used to restore archived images from picture archiving and communication systems to diagnostic imaging workstations prior to anticipated need, facilitating timely comparison of historical studies and patient management. The authors describe a problem-oriented prefetching scheme, detailing 1) a mechanism supporting selection of patients for prefetching via characterizations of clinical problems, using multiple data sources (picture archiving and communication systems, hospital information systems, and radiology information systems), classifying patients into cohorts on the basis of their medical conditions (e.g., lung cancer); and 2) prefetching of multimedia data (imaging, laboratory, and medical reports) from clinical databases to enable the viewing of an integrated patient record. Preliminary evaluation of the prefetching algorithm using classic information retrieval measures showed that the system had high recall (100 percent), correctly identifying and retrieving data for all patients belonging to a target cohort, but low precision (50 percent). A key finding during testing was that the recall of the system was increased through the use of multiple data sources (compared with one data source), because of better patient descriptors. Medical problems and patient cohorts were more specifically defined by combining information from heterogeneous databases.

The purpose of this work is to develop patient-specific models for automatically detecting lung nodules in computed tomography (CT) images. It is motivated by significant developments in CT scanner technology and the burden that lung cancer screening and surveillance imposes on radiologists. We propose a new method that uses a patient's baseline image data to assist in the segmentation of subsequent images so that changes in size and/or shape of nodules can be measured automatically. The system uses a generic, a priori model to detect candidate nodules on the baseline scan of a previously unseen patient. A user then confirms or rejects nodule candidates to establish baseline results. For analysis of follow-up scans of that particular patient, a patient-specific model is derived from these baseline results. This model describes expected features (location, volume and shape) of previously segmented nodules so that the system can relocalize them automatically on follow-up. On the baseline scans of 17 subjects, a radiologist identified a total of 36 nodules, of which 31 (86%) were detected automatically by the system with an average of 11 false positives (FPs) per case. In follow-up scans 27 of the 31 nodules were still present and, using patient-specific models, 22 (81%) were correctly relocalized by the system. The system automatically detected 16 out of a possible 20 (80%) of new nodules on follow-up scans with ten FPs per case.

A knowledge-based system has been developed for segmenting computed tomography (CT) images. The modular architecture includes an anatomical model, image processing engine, inference engine and blackboard. The model contains a priori knowledge of size, shape, X-ray attenuation and relative position of anatomical structures. This knowledge is used to constrain low-level segmentation routines. Model-derived constraints and segmented image objects are both transformed into a common feature space and posted on the blackboard. The inference engine then matches image to model objects, based on the constraints. The transformation to feature space allows the knowledge and image data representations to be independent. Thus a high-level model can be used, with data being stored in a frame-based semantic network. Also, standard, low-level segmentation algorithms can be easily plugged-in to the system. This modularity allows for straightforward system extension and maintenance. We initially demonstrate an application to lung segmentation in thoracic CT, with subsequent extension of the knowledge-base to include tumors within the lung fields.

We present an automated, knowledge-based method for segmenting chest computed tomography (CT) datasets. Anatomical knowledge including expected volume, shape, relative position, and X-ray attenuation of organs provides feature constraints that guide the segmentation process. Knowledge is represented at a high level using an explicit anatomical model. The model is stored in a frame-based semantic network and anatomical variability is incorporated using fuzzy sets. A blackboard architecture permits the data representation and processing algorithms in the model domain to be independent of those in the image domain. Knowledge-constrained segmentation routines extract contiguous three-dimensional (3-D) sets of voxels, and their feature-space representations are posted on the blackboard. An inference engine uses fuzzy logic to match image to model objects based on the feature constraints. Strict separation of model and image domains allows for systematic extension of the knowledge base. In preliminary experiments, the method has been applied to a small number of thoracic CT datasets. Based on subjective visual assessment by experienced thoracic radiologists, basic anatomic structures such as the lungs, central tracheobronchial tree, chest wall, and mediastinum were successfully segmented. To demonstrate the extensibility of the system, knowledge was added to represent the more complex anatomy of lung lesions in contact with vessels or the chest wall. Visual inspection of these segmented lesions was also favorable. These preliminary results suggest that use of expert knowledge provides an increased level of automation compared with low-level segmentation techniques. Moreover, the knowledge-based approach may better discriminate between structures of similar attenuation and anatomic contiguity. Further validation is required.