Automated detection of lung nodules in CT scans: preliminary results

Pulmonary nodules on multi-detector row CT scans: performance comparison of radiologists and computer-aided detection

PURPOSE

To compare the performance of radiologists and of a computer-aided detection (CAD) algorithm for pulmonary nodule detection on thin-section thoracic computed tomographic (CT) scans.

MATERIALS AND METHODS

The study was approved by the institutional review board. The requirement of informed consent was waived. Twenty outpatients (age range, 15-91 years; mean, 64 years) were examined with chest CT (multi-detector row scanner, four detector rows, 1.25-mm section thickness, and 0.6-mm interval) for pulmonary nodules. Three radiologists independently analyzed CT scans, recorded the locus of each nodule candidate, and assigned each a confidence score. A CAD algorithm with parameters chosen by using cross validation was applied to the 20 scans. The reference standard was established by two experienced thoracic radiologists in consensus, with blind review of all nodule candidates and free search for additional nodules at a dedicated workstation for three-dimensional image analysis. True-positive (TP) and false-positive (FP) results and confidence levels were used to generate free-response receiver operating characteristic (ROC) plots. Double-reading performance was determined on the basis of TP detections by either reader.

RESULTS

The 20 scans showed 195 noncalcified nodules with a diameter of 3 mm or more (reference reading). Area under the alternative free-response ROC curve was 0.54, 0.48, 0.55, and 0.36 for CAD and readers 1-3, respectively. Differences between reader 3 and CAD and between readers 2 and 3 were significant (P < .05); those between CAD and readers 1 and 2 were not significant. Mean sensitivity for individual readings was 50% (range, 41%-60%); double reading resulted in increase to 63% (range, 56%-67%). With CAD used at a threshold allowing only three FP detections per CT scan, mean sensitivity was increased to 76% (range, 73%-78%). CAD complemented individual readers by detecting additional nodules more effectively than did a second reader; CAD-reader weighted kappa values were significantly lower than reader-reader weighted kappa values (Wilcoxon rank sum test, P < .05).

CONCLUSION

With CAD used at a level allowing only three FP detections per CT scan, sensitivity was substantially higher than with conventional double reading.

Volumetric Assessment of Pulmonary Nodules with ECG-Gated MDCT

OBJECTIVE

The objective of our study was to assess physiologic lung deformation and compression originating from cardiovascular motion and their subsequent impact on determining the volume of small pulmonary nodules throughout the cardiac cycle on ECG-gated MDCT.

SUBJECTS AND METHODS

Seventy-three small noncalcified pulmonary nodules were identified in 30 patients who underwent ECG-gated MDCT. The volume of each nodule was assessed throughout the cardiac cycle using computer-aided automatic segmentation algorithms, and the assessment was repeated three times. To ensure the validity of the subtle changes in volume that were detected, we determined the volume and signal attenuation in phantom data sets and patient nodules without temporal or spatial differentiation. Subsequently, nodules were assigned to pulmonary segments, and volume changes were correlated to cardiac phases, nodular location, and mean nodular size. Statistical multivariate tests were performed to evaluate significant patterns.

RESULTS

The validity of significant measurements was proven in evaluated phantom data sets with a general tendency toward overestimating nodular volume (p = 0.492). Statistical evaluation of nodular signal attenuation confirmed true deformation and compression of nodules rather than partial volume effects as the reason for volume variations (p = 0.874). Differentiating pulmonary nodules in cardiac phases, pulmonary locations, and mean nodular volumes, we found that one single effect did not determine the amount of cardiovascular motion conveyed to pulmonary parenchyma and subsequently led to nodule deformation. Multivariate testing revealed statistically significant measures identifying patterns correlating variation in nodular volume with cardiac phase (p < 0.001), nodular location (p = 0.007), and mean nodular size (p < 0.001).

CONCLUSION

Cardiovascular motion was disproportionately conveyed to various pulmonary segments and led to changes in the volume of pulmonary nodules, especially in small pulmonary nodules. A precise volumetric assessment was therefore possible only by identifying the underlying cardiac phase.

Lung image database consortium: developing a resource for the medical imaging research community

To stimulate the advancement of computer-aided diagnostic (CAD) research for lung nodules in thoracic computed tomography (CT), the National Cancer Institute launched a cooperative effort known as the Lung Image Database Consortium (LIDC). The LIDC is composed of five academic institutions from across the United States that are working together to develop an image database that will serve as an international research resource for the development, training, and evaluation of CAD methods in the detection of lung nodules on CT scans. Prior to the collection of CT images and associated patient data, the LIDC has been engaged in a consensus process to identify, address, and resolve a host of challenging technical and clinical issues to provide a solid foundation for a scientifically robust database. These issues include the establishment of (a) a governing mission statement, (b) criteria to determine whether a CT scan is eligible for inclusion in the database, (c) an appropriate definition of the term qualifying nodule, (d) an appropriate definition of "truth" requirements, (e) a process model through which the database will be populated, and (f) a statistical framework to guide the application of assessment methods by users of the database. Through a consensus process in which careful planning and proper consideration of fundamental issues have been emphasized, the LIDC database is expected to provide a powerful resource for the medical imaging research community. This article is intended to share with the community the breadth and depth of these key issues.

Automated lung segmentation for thoracic CT impact on computer-aided diagnosis

RATIONALE AND OBJECTIVES

Automated lung segmentation in thoracic computed tomography scans is essential for the development of computer-aided diagnostic (CAD) methods. A core segmentation method may be developed for general application; however, modifications may be required for specific clinical tasks.

MATERIALS AND METHODS

An automated lung segmentation method has been applied (1) as preprocessing for automated lung nodule detection and (2) as the foundation for computer-assisted measurements of pleural mesothelioma tumor thickness. The core method uses gray-level thresholding to segment the lungs within each computed tomography section. The segmentation is revised through separation of right and left lungs along the anterior junction line, elimination of the trachea and main bronchi from the lung segmentation regions, and suppression of the diaphragm. Segmentation modifications required for nodule detection include a rolling ball algorithm to include juxtapleural nodules and morphologic erosion to eliminate partial volume pixels at the boundary of the segmentation regions.

RESULTS

For automated lung nodule detection, 4 of 82 actual nodules (4.9%) were excluded from the lung segmentation regions when the core segmentation method was modified compared with 14 nodules (17.1%) excluded without modifications. The computer-assisted quantification of mesothelioma method achieved a correlation coefficient of 0.990 with 134 manual measurements when the core segmentation method was used alone; correlation was reduced to 0.977 when the segmentation modifications, as adapted for the lung nodule detection task, were applied to the mesothelioma measurement task.

CONCLUSION

Different CAD applications impose different requirements on the automated lung segmentation process. The specific approach to lung segmentation must be adapted to the particular CAD task.

Computer-aided diagnosis in high resolution CT of the lungs

A computer-aided diagnosis (CAD) system is presented to automatically distinguish normal from abnormal tissue in high-resolution CT chest scans acquired during daily clinical practice. From high-resolution computed tomography scans of 116 patients, 657 regions of interest are extracted that are to be classified as displaying either normal or abnormal lung tissue. A principled texture analysis approach is used, extracting features to describe local image structure by means of a multi-scale filter bank. The use of various classifiers and feature subsets is compared and results are evaluated with ROC analysis. Performance of the system is shown to approach that of two expert radiologists in diagnosing the local regions of interest, with an area under the ROC curve of 0.862 for the CAD scheme versus 0.877 and 0.893 for the radiologists.

Computer-assisted detection of pulmonary nodules: performance evaluation of an expert knowledge-based detection system in consensus reading with experienced and inexperienced chest radiologists

To evaluate the performance of experienced versus inexperienced radiologists in comparison and in consensus with an interactive computer-aided detection (CAD) system for detection of pulmonary nodules. Eighteen consecutive patients (mean age: 62.2 years; range 29-83 years) prospectively underwent routine 16-row multislice computed tomography (MSCT). Four blinded radiologists (experienced: readers 1, 2; inexperienced: readers 3, 4) assessed image data against CAD for pulmonary nodules. Thereafter, consensus readings of readers 1+3, reader 1+CAD and reader 3+CAD were performed. Data were compared against an independent gold standard. Statistical tests used to calculate interobserver agreement, reader performance and nodule size were Kappa, ROC and Mann-Whitney U. CAD and experienced readers outperformed inexperienced readers (Az=0.72, 0.71, 0.73, 0.49 and 0.50 for CAD, readers 1-4, respectively; P<0.05). Performance of reader 1+CAD was superior to single reader and reader 1+3 performances (Az=0.93, 0.72 for reader 1+CAD and reader 1+3 consensus, respectively, P<0.05). Reader 3+CAD did not perform superiorly to experienced readers or CAD (Az=0.79 for reader 3+CAD; P>0.05). Consensus of reader 1+CAD significantly outperformed all other readings, demonstrating a benefit in using CAD as an inexperienced reader replacement. It is questionable whether inexperienced readers can be regarded as adequate for interpretation of pulmonary nodules in consensus with CAD, replacing an experienced radiologist.

Population screening for lung cancer

The feasibility of diagnosing small stage 1 lung cancers using low-dose chest computed tomography in asymptomatic at-risk individuals has been demonstrated in multiple studies. However, it has yet to be proved that the introduction of a chest computed tomography screening programme would do more good than harm at an acceptable cost.

Automated lung nodule detection at low-dose CT: preliminary experience

Objective

To determine the usefulness of a computer-aided diagnosis (CAD) system for the automated detection of lung nodules at low-dose CT.

Materials and Methods

A CAD system developed for detecting lung nodules was used to process the data provided by 50 consecutive low-dose CT scans. The results of an initial report, a second look review by two chest radiologists, and those obtained by the CAD system were compared, and by reviewing all of these, a gold standard was established.

Results

By applying the gold standard, a total of 52 nodules were identified (26 with a diameter ≤ 5 mm; 26 with a diameter > 5 mm). Compared to an initial report, four additional nodules were detected by the CAD system. Three of these, identified only at CAD, formed part of the data used to derive the gold standard. For the detection of nodules > 5 mm in diameter, sensitivity was 77% for the initial report, 88% for the second look review, and 65% for the CAD system. There were 8.0±5.2 false-positive CAD results per CT study.

Conclusion

These preliminary results indicate that a CAD system may improve the detection of pulmonary nodules at low-dose CT.

Surface normal overlap: a computer-aided detection algorithm with application to colonic polyps and lung nodules in helical CT

We developed a novel computer-aided detection (CAD) algorithm called the surface normal overlap method that we applied to colonic polyp detection and lung nodule detection in helical computed tomography (CT) images. We demonstrate some of the theoretical aspects of this algorithm using a statistical shape model. The algorithm was then optimized on simulated CT data and evaluated using a per-lesion cross-validation on 8 CT colonography datasets and on 8 chest CT datasets. It is able to achieve 100% sensitivity for colonic polyps 10 mm and larger at 7.0 false positives (FPs)/dataset and 90% sensitivity for solid lung nodules 6 mm and larger at 5.6 FP/dataset.

Computerized scheme for automated detection of lung nodules in low-dose computed tomography images for lung cancer screening

RATIONALE AND OBJECTIVES

A computerized scheme for automated detection of lung nodules in low-dose computed tomography images for lung cancer screening was developed.

MATERIALS AND METHODS

Our scheme is based on a difference-image technique for enhancing the lung nodules and suppressing the majority of background normal structures. The difference image for each computed tomography image was obtained by subtracting the nodule-suppressed image processed with a ring average filter from the nodule-enhanced image with a matched filter. The initial nodule candidates were identified by applying a multiple-gray level thresholding technique to the difference image, where most nodules were well enhanced. A number of false-positives were removed first in entire lung regions and second in divided lung regions by use of the two rule-based schemes on the localized image features related to morphology and gray levels. Some of the remaining false-positives were eliminated by use of a multiple massive training artificial neural network trained for reduction of various types of false-positives. This computerized scheme was applied to a confirmed cancer database of 106 low-dose computed tomography scans with 109 cancer lesions for 73 patients obtained from a lung cancer screening program in Nagano, Japan.

RESULTS

This computed-aided diagnosis scheme provided a sensitivity of 83% (91/109) for all cancers with 5.8 false-positives per scan, which included 84% (32/38) for missed cancers with 5.9 false-positives per scan.

CONCLUSION

This computerized scheme may be useful for assisting radiologists in detecting lung cancers on low-dose computed tomography images for lung cancer screening.

Automated lung segmentation in X-ray computed tomography: development and evaluation of a heuristic threshold-based scheme

RATIONALE AND OBJECTIVES

To develop and evaluate a reliable, fully-automated lung segmentation scheme for application in X-ray computed tomography.

MATERIALS AND METHODS

The automated scheme was heuristically developed using a slice-based, pixel-value threshold and two sets of classification rules. Features used in the rules include size, circularity, and location. The segmentation scheme operates slice-by-slice and performs three key operations: (1) image preprocessing to remove background pixels, (2) computation and application of a pixel-value threshold to identify lung tissue, and (3) refinement of the initial segmented regions to prune incorrectly detected airways and separate fused right and left lungs.

RESULTS

The performance of the automated segmentation scheme was evaluated using 101 computed tomography cases (91 thick slice, 10 thin slice scans). The 91 thick cases were pre- and post-surgery from 50 patients and were not independent. The automated scheme successfully segmented 94.0% of the 2,969 thick slice images and 97.6% of the 1,161 thin slice images. The mean difference of the total lung volumes calculated by the automated scheme and functional residual capacity plus 60% inspiratory capacity was -24.7 +/- 508.1 mL. The mean differences of the total lung volumes calculated by the automated scheme and an established, commonly used semi-automated scheme were 95.2 +/- 52.5 mL and -27.7 +/- 66.9 mL for the thick and thin slice cases, respectively.

CONCLUSION

This simple, fully-automated lung segmentation scheme provides an objective tool to facilitate lung segmentation from computed tomography scans.

Diagnostic fidelity of the Z-buffer segmentation algorithm: preliminary assessment based on intracranial aneurysm detection

We have developed an algorithm known as the Z-buffer segmentation (ZBS) algorithm for segmenting vascular structures from 3D MRA images. Previously we evaluated the accuracy of the ZBS algorithm on a voxel level in terms of inclusion and exclusion of vascular and background voxels. In this paper we evaluate the diagnostic fidelity of the ZBS algorithm. By diagnostic fidelity we mean that the data preserves the structural information necessary for diagnostic evaluation. This evaluation is necessary to establish the potential usefulness of the segmentation for improved image display, or whether the segmented data could form the basis of a computerized analysis tool. We assessed diagnostic fidelity by measuring how well human observers could detect aneurysms in the segmented data sets. ZBS segmentation of 30 MRA cases containing 29 aneurysms was performed. Image display used densitometric reprojections with shaded surface highlighting that were generated from the segmented data. Three neuroradiologists independently reviewed the generated ZBS images for aneurysms. The observers had 80% sensitivity (90% for aneurysms larger than 2mm) with 0.13 false positives per image. Good agreement with the gold standard for describing aneurysm size and orientation was shown. These preliminary results suggest that the segmentation has diagnostic fidelity with the original data and may be useful for improved visualization or automated analysis of the vasculature.

Future generation CT imaging

X-ray CT technology has been available for more than 30 years, yet continued technological advances have kept CT imaging at the forefront of medical imaging innovation. Consequently, the number of clinical CT applications has increased steadily. Other imaging modalities might be superior to CT imaging for some specific applications, but no other single modality is more often used in chest imaging today. Future technological developments in the area of high-resolution detectors, high-capacity x-ray tubes, advanced reconstruction algorithms, and improved visualization techniques will continue to expand the imaging capability. Future CT imaging technology will combine improved imaging capability with advanced and specific computer-assisted tools, which will expand the usefulness of CT imaging in many areas.

Image annotation for conveying automated lung nodule detection results to radiologists

RATIONALE AND OBJECTIVES

The author investigated the ability of automated techniques to convey the results of an automated lung nodule detection method for human visualization.

MATERIALS AND METHODS

Automated nodule detection begins with gray-level thresholding techniques to create a segmented lung volume within which nodule candidates are identified. Morphologic and gray-level features are computed for each candidate. To distinguish between candidates that represent actual nodules and those that represent non-nodules, a rule-based scheme is combined with linear discriminant analysis. For output visualization, final detection results are represented as circles around computer-detected structures in a single section in which each structure appears. Consequently, an inappropriate choice of section could result in an actual nodule detected by the computer but not properly indicated to the radiologist, thus reducing the potential positive impact of that detection on the radiologist's decision-making process.

RESULTS

The automated nodule detection method achieved 71% sensitivity with 0.5 false positives per section on 38 CT scans; however, when these results were converted to annotations on the images output for human visualization, only 91% of the computer-detected true-positive nodules received annotations that encompassed a portion of the actual nodule. Thus, the "effective sensitivity" of the automated detection method was reduced.

CONCLUSION

The "effective sensitivity" of an automated lung nodule detection system considers the eventual human interaction with system output. Differences between reported computer sensitivity and "effective sensitivity" may be reduced through proper consideration of the assessment of "truth," of the manner in which computer results are scored, and of the complete segmentation of candidates for automated nodule detection.

Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm

Increasingly, computed tomography (CT) offers higher resolution and faster acquisition times. This has resulted in the opportunity to detect small lung nodules, which may represent lung cancers at earlier and potentially more curable stages. However, in the current clinical practice, hundreds of such thin-sectional CT images are generated for each patient and are evaluated by a radiologist in the traditional sense of looking at each image in the axial mode. This results in the potential to miss small nodules and thus potentially miss a cancer. In this paper, we present a computerized method for automated identification of small lung nodules on multislice CT (MSCT) images. The method consists of three steps: (i) separation of the lungs from the other anatomic structures, (ii) detection of nodule candidates in the extracted lungs, and (iii) reduction of false-positives among the detected nodule candidates. A three-dimensional lung mask can be extracted by analyzing density histogram of volumetric chest images followed by a morphological operation. Higher density structures including nodules scattered throughout the lungs can be identified by using a local density maximum algorithm. Information about nodules such as size and compact shape are then incorporated into the algorithm to reduce the detected nodule candidates which are not likely to be nodules. The method was applied to the detection of computer simulated small lung nodules (2 to 7 mm in diameter) and achieved a sensitivity of 84.2% with, on average, five false-positive results per scan. The preliminary results demonstrate the potential of this technique for assisting the detection of small nodules from chest MSCT images.

Massive training artificial neural network (MTANN) for reduction of false positives in computerized detection of lung nodules in low-dose computed tomography

In this study, we investigated a pattern-recognition technique based on an artificial neural network (ANN), which is called a massive training artificial neural network (MTANN), for reduction of false positives in computerized detection of lung nodules in low-dose computed tomography (CT) images. The MTANN consists of a modified multilayer ANN, which is capable of operating on image data directly. The MTANN is trained by use of a large number of subregions extracted from input images together with the teacher images containing the distribution for the "likelihood of being a nodule." The output image is obtained by scanning an input image with the MTANN. The distinction between a nodule and a non-nodule is made by use of a score which is defined from the output image of the trained MTANN. In order to eliminate various types of non-nodules, we extended the capability of a single MTANN, and developed a multiple MTANN (Multi-MTANN). The Multi-MTANN consists of plural MTANNs that are arranged in parallel. Each MTANN is trained by using the same nodules, but with a different type of non-nodule. Each MTANN acts as an expert for a specific type of non-nodule, e.g., five different MTANNs were trained to distinguish nodules from various-sized vessels; four other MTANNs were applied to eliminate some other opacities. The outputs of the MTANNs were combined by using the logical AND operation such that each of the trained MTANNs eliminated none of the nodules, but removed the specific type of non-nodule with which the MTANN was trained, and thus removed various types of non-nodules. The Multi-MTANN consisting of nine MTANNs was trained with 10 typical nodules and 10 non-nodules representing each of nine different non-nodule types (90 training non-nodules overall) in a training set. The trained Multi-MTANN was applied to the reduction of false positives reported by our current computerized scheme for lung nodule detection based on a database of 63 low-dose CT scans (1765 sections), which contained 71 confirmed nodules including 66 biopsy-confirmed primary cancers, from a lung cancer screening program. The Multi-MTANN was applied to 58 true positives (nodules from 54 patients) and 1726 false positives (non-nodules) reported by our current scheme in a validation test; these were different from the training set. The results indicated that 83% (1424/1726) of non-nodules were removed with a reduction of one true positive (nodule), i.e., a classification sensitivity of 98.3% (57 of 58 nodules). By using the Multi-MTANN, the false-positive rate of our current scheme was improved from 0.98 to 0.18 false positives per section (from 27.4 to 4.8 per patient) at an overall sensitivity of 80.3% (57/71).

Automated lung nodule classification following automated nodule detection on CT: a serial approach

We have evaluated the performance of an automated classifier applied to the task of differentiating malignant and benign lung nodules in low-dose helical computed tomography (CT) scans acquired as part of a lung cancer screening program. The nodules classified in this manner were initially identified by our automated lung nodule detection method, so that the output of automated lung nodule detection was used as input to automated lung nodule classification. This study begins to narrow the distinction between the "detection task" and the "classification task." Automated lung nodule detection is based on two- and three-dimensional analyses of the CT image data. Gray-level-thresholding techniques are used to identify initial lung nodule candidates, for which morphological and gray-level features are computed. A rule-based approach is applied to reduce the number of nodule candidates that correspond to non-nodules, and the features of remaining candidates are merged through linear discriminant analysis to obtain final detection results. Automated lung nodule classification merges the features of the lung nodule candidates identified by the detection algorithm that correspond to actual nodules through another linear discriminant classifier to distinguish between malignant and benign nodules. The automated classification method was applied to the computerized detection results obtained from a database of 393 low-dose thoracic CT scans containing 470 confirmed lung nodules (69 malignant and 401 benign nodules). Receiver operating characteristic (ROC) analysis was used to evaluate the ability of the classifier to differentiate between nodule candidates that correspond to malignant nodules and nodule candidates that correspond to benign lesions. The area under the ROC curve for this classification task attained a value of 0.79 during a leave-one-out evaluation.

Lung nodule detection and characterization with multislice CT

The ability to identify and characterize pulmonary nodules has been dramatically increased by the introduction of multislice CT (MSCT) technology. Using high-resolution sections, MSCT allows considerable improvement in assessing nodule morphology, enhancement patterns, and growth. MSCT also has facilitated the development and potential of clinical application of computer-assisted diagnosis.

Alignment of CT lung volumes with an optical flow method

RATIONALE AND OBJECTIVES

This study was performed to evaluate an optical flow method for registering serial computed tomographic (CT) images of lung volumes to assist physicians in visualizing and assessing changes between CT scans.

MATERIALS AND METHODS

The optical flow method is a coarse-to-fine model-based motion estimation technique for estimating first a global parametric transformation and then local deformations. Five serial pairs of CT images of lung volumes that were misaligned because of patient positioning, respiration, and/or different fields of view were used to test the method.

RESULTS

Lung volumes depicted on the serial paired images initially were correlated at only 28%-68% because of misalignment. With use of the optical flow method, the serial images were aligned to at least 95% correlation.

CONCLUSION

The optical flow method enables a direct comparison of serial CT images of lung volumes for the assessment of nodules or functional changes in the lung.

Automated detection of lung nodules in CT scans: effect of image reconstruction algorithm

We have investigated the effect of computed tomography (CT) image reconstruction algorithm on the performance of our automated lung nodule detection method. Commercial CT scanners offer a choice of several algorithms for the reconstruction of projection data into transaxial images. Different algorithms produce images with substantially different properties that are apparent not only quantitatively, but also through visual assessment. During some clinical thoracic CT examinations, patient scans are reconstructed with multiple reconstruction algorithms. Thirty-eight such cases were collected to form two databases: one with patient projection data reconstructed with the "standard" reconstruction algorithm and the other with the same patient projection data reconstructed with the "lung" reconstruction algorithm. The automated nodule detection method was applied to both databases. This method is based on gray-level-thresholding techniques to segment the lung regions from each CT section to create a segmented lung volume. Further gray-level-thresholding techniques are applied within the segmented lung volume to identify a set of lung nodule candidates. Rule-based and linear discriminant classifiers are used to differentiate between lung nodule candidates that correspond to actual nodules and those that correspond to non-nodules. The automated method that was applied to both databases was exactly the same, except that the classifiers were calibrated separately for each database. For comparison, the classifier then was trained on one database and tested independently on the other database. When applied to the databases in this manner, the automated method demonstrated overall a similar level of performance, indicating an encouraging degree of robustness.

Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm

Increasingly, computed tomography (CT) offers higher resolution and faster acquisition times. This has resulted in the opportunity to detect small lung nodules, which may represent lung cancers at earlier and potentially more curable stages. However, in the current clinical practice, hundreds of such thin-sectional CT images are generated for each patient and are evaluated by a radiologist in the traditional sense of looking at each image in the axial mode. This results in the potential to miss small nodules and thus potentially miss a cancer. In this paper, we present a computerized method for automated identification of small lung nodules on multislice CT (MSCT) images. The method consists of three steps: (i) separation of the lungs from the other anatomic structures, (ii) detection of nodule candidates in the extracted lungs, and (iii) reduction of false-positives among the detected nodule candidates. A three-dimensional lung mask can be extracted by analyzing density histogram of volumetric chest images followed by a morphological operation. Higher density structures including nodules scattered throughout the lungs can be identified by using a local density maximum algorithm. Information about nodules such as size and compact shape are then incorporated into the algorithm to reduce the detected nodule candidates which are not likely to be nodules. The method was applied to the detection of computer simulated small lung nodules (2 to 7 mm in diameter) and achieved a sensitivity of 84.2% with, on average, five false-positive results per scan. The preliminary results demonstrate the potential of this technique for assisting the detection of small nodules from chest MSCT images.

Maximum likelihood fitting of FROC curves under an initial-detection-and-candidate-analysis model

We have developed a model for FROC curve fitting that relates the observer's FROC performance not to the ROC performance that would be obtained if the observer's responses were scored on a per image basis, but rather to a hypothesized ROC performance that the observer would obtain in the task of classifying a set of "candidate detections" as positive or negative. We adopt the assumptions of the Bunch FROC model, namely that the observer's detections are all mutually independent, as well as assumptions qualitatively similar to, but different in nature from, those made by Chakraborty in his AFROC scoring methodology. Under the assumptions of our model, we show that the observer's FROC performance is a linearly scaled version of the candidate analysis ROC curve, where the scaling factors are just given by the FROC operating point coordinates for detecting initial candidates. Further, we show that the likelihood function of the model parameters given observational data takes on a simple form, and we develop a maximum likelihood method for fitting a FROC curve to this data. FROC and AFROC curves are produced for computer vision observer datasets and compared with the results of the AFROC scoring method. Although developed primarily with computer vision schemes in mind, we hope that the methodology presented here will prove worthy of further study in other applications as well.

Lung cancers missed at low-dose helical CT screening in a general population: comparison of clinical, histopathologic, and imaging findings

PURPOSE

To compare clinical, histopathologic, and imaging features of lung cancers missed at low-radiation-dose helical computed tomography (CT).

MATERIALS AND METHODS

Eighty-three primary lung cancers were found during an annual low-dose CT screening program and confirmed histopathologically at either surgery or biopsy. Thirty-two of these lung cancers were missed on 39 CT scans: on 23 scans owing to detection errors and on 16 owing to interpretation errors. The clinical characteristics, CT features, and histopathologic findings of these missed lung cancers were correlated.

RESULTS

All missed cancers were intrapulmonary, and 28 (88%) were stage IA. All 20 detection errors occurred in cases of adenocarcinoma, 17 (85%) of which were well-differentiated tumors and 11 (55%) of which were in nonsmoking women. The mean size of cancers missed owing to detection error, 9.8 mm, was smaller than that of cancers missed owing to interpretation error, 15.9 mm (P <.001). In the detection error group, the percentages of nodules with ground-glass opacity (91%) or judged to be subtle (91%) were greater than those of nodules in the interpretation error group (38% and 25%, respectively) (P <.001). In the detection error group, 83% (19/23) of cancers were overlapped with, obscured by, or similar in appearance to normal structures such as pulmonary vessels. On 14 of the 16 CT scans with which there were interpretation errors, the CT findings mimicked benign disease, and the patients also had underlying lung disease, such as tuberculosis, emphysema, or lung fibrosis.

CONCLUSION

The lung cancers missed at low-dose CT screening in this series generally were very subtle and appeared as small faint nodules, overlapping normal structures, or opacities in a complex background of other disease.

Lung cancer: performance of automated lung nodule detection applied to cancers missed in a CT screening program

PURPOSE

To evaluate the performance of a fully automated computerized method for the detection of lung nodules in computed tomographic (CT) scans in the identification of lung cancers that may be missed during visual interpretation.

MATERIALS AND METHODS

A database of 38 low-dose CT scans with 50 lung nodules was obtained from a lung cancer screening program. Thirty-eight of the nodules represented biopsy-confirmed lung cancers that had not been reported during initial clinical interpretation. A computer detection method that involved the use of gray-level thresholding techniques to identify three-dimensionally contiguous structures within the lungs was applied to the CT data. Computer-extracted volume was used to determine whether a structure became a nodule candidate. A rule-based scheme and a cascaded automated classifier were applied to the set of nodule candidates to distinguish actual nodules from areas of normal anatomy. Overall performance of the computer detection method was evaluated with free-response receiver operating characteristic (FROC) analysis.

RESULTS

At a specific operating point on the FROC curve, the method achieved a sensitivity of 80% (40 of 50 nodules), with an average of 1.0 false-positive detection per section. Missed cancers were detected by the computerized method with a sensitivity of 84% (32 of 38 nodules) and a false-positive rate of 1.0 per section.

CONCLUSION

With an automated lung nodule detection method, a large fraction (84%, 32 of 38) of missed cancers in a database of low-dose CT scans were detected correctly.

Lung nodule detection on thoracic computed tomography images: preliminary evaluation of a computer-aided diagnosis system

We are developing a computer-aided diagnosis (CAD) system for lung nodule detection on thoracic helical computed tomography (CT) images. In the first stage of this CAD system, lung regions are identified by a k-means clustering technique. Each lung slice is classified as belonging to the upper, middle, or the lower part of the lung volume. Within each lung region, structures are segmented again using weighted k-means clustering. These structures may include true lung nodules and normal structures consisting mainly of blood vessels. Rule-based classifiers are designed to distinguish nodules and normal structures using 2D and 3D features. After rule-based classification, linear discriminant analysis (LDA) is used to further reduce the number of false positive (FP) objects. We performed a preliminary study using 1454 CT slices from 34 patients with 63 lung nodules. When only LDA classification was applied to the segmented objects, the sensitivity was 84% (53/63) with 5.48 (7961/1454) FP objects per slice. When rule-based classification was used before LDA, the free response receiver operating characteristic (FROC) curve improved over the entire sensitivity and specificity ranges of interest. In particular, the FP rate decreased to 1.74 (2530/1454) objects per slice at the same sensitivity. Thus, compared to FP reduction with LDA alone, the inclusion of rule-based classification lead to an improvement in detection accuracy for the CAD system. These preliminary results demonstrate the feasibility of our approach to lung nodule detection and FP reduction on CT images.