Reduction of false positives in computerized detection of lung nodules in chest radiographs using artificial neural networks, discriminant analysis, and a rule-based scheme

3D deep learning for detecting pulmonary nodules in CT scans

Objective

To demonstrate and test the validity of a novel deep-learning-based system for the automated detection of pulmonary nodules.

Materials and Methods

The proposed system uses 2 3D deep learning models, 1 for each of the essential tasks of computer-aided nodule detection: candidate generation and false positive reduction. A total of 888 scans from the LIDC-IDRI dataset were used for training and evaluation.

Results

Results for candidate generation on the test data indicated a detection rate of 94.77% with 30.39 false positives per scan, while the test results for false positive reduction exhibited a sensitivity of 94.21% with 1.789 false positives per scan. The overall system detection rate on the test data was 89.29% with 1.789 false positives per scan.

Discussion

An extensive and rigorous validation was conducted to assess the performance of the proposed system. The system demonstrated a novel combination of 3D deep neural network architectures and demonstrates the use of deep learning for both candidate generation and false positive reduction to be evaluated with a substantial test dataset. The results strongly support the ability of deep learning pulmonary nodule detection systems to generalize to unseen data. The source code and trained model weights have been made available.

Conclusion

A novel deep-neural-network-based pulmonary nodule detection system is demonstrated and validated. The results provide comparison of the proposed deep-learning-based system over other similar systems based on performance.

Review: On Segmentation of Nodules from Posterior and Anterior Chest Radiographs

Lung cancer is one of the major types of cancer in the world. Survival rate can be increased if the disease can be identified early. Posterior and anterior chest radiography and computerized tomography scans are the most used diagnosis techniques for detecting tumor from lungs. Posterior and anterior chest radiography requires less radiation dose and is available in most of the diagnostic centers and it costs less compared to the remaining diagnosis techniques. So PA chest radiography became the most commonly used technique for lung cancer detection. Because of superimposed anatomical structures present in the image, sometimes radiologists cannot find abnormalities from the image. To help radiologists in diagnosing tumor from PA chest radiographic images range of CAD scheme has been developed for the past three decades. These computerized tools may be used by radiologists as a second opinion in detecting tumor. Literature survey on detecting tumors from chest graphs is presented in this paper.

A novel approach to CAD system for the detection of lung nodules in CT images

Detection of pulmonary nodule plays a significant role in the diagnosis of lung cancer in early stage that improves the chances of survival of an individual. In this paper, a computer aided nodule detection method is proposed for the segmentation and detection of challenging nodules like juxtavascular and juxtapleural nodules. Lungs are segmented from computed tomography (CT) images using intensity thresholding; brief analysis of CT image histogram is done to select a suitable threshold value for better segmentation results. Simple morphological closing is used to include juxtapleural nodules in segmented lung regions. K-means clustering is applied for the initial detection and segmentation of potential nodules; shape specific morphological opening is implemented to refine segmentation outcomes. These segmented potential nodules are then divided into six groups on the basis of their thickness and percentage connectivity with lung walls. Grouping not only helped in improving system's efficiency but also reduced computational time, otherwise consumed in calculating and analyzing unnecessary features for all nodules. Different sets of 2D and 3D features are extracted from nodules in each group to eliminate false positives. Small size nodules are differentiated from false positives (FPs) on the basis of their salient features; sensitivity of the system for small nodules is 83.33%. SVM classifier is used for the classification of large nodules, for which the sensitivity of the proposed system is 93.8% applying 10-fold cross-validation. Receiver Operating Characteristic (ROC) curve is used for the analysis of CAD system. Overall sensitivity of the system is 91.65% with 3.19 FPs per case, and accuracy is 96.22%. The system took 3.8 seconds to analyze each image.

Computer-Aided Diagnosis for Chest Radiographs in Intensive Care

The chest radiograph is an essential tool for the diagnosis of several lung diseases in intensive care units (ICU). However, several factors make the interpretation of the chest radiograph difficult including the number of X-rays done daily in ICU, the quality of the chest radiograph, and the lack of a standardized interpretation. To overcome these limitations in the interpretation of chest radiographs, researchers have developed computer-aided diagnosis (CAD) systems. In this review, the authors report the methodology used to develop CAD systems including identification of the region of interest, analysis of these regions, and classification. Currently, only a few CAD systems for chest X-ray interpretation are commercially available. Some promising research is ongoing, but the involvement of the pediatric research community is needed for the development and validation of such CAD systems dedicated to pediatric intensive care.

Development of a computer-aided diagnostic scheme for detection of interval changes in successive whole-body bone scans

Bone scintigraphy is the most frequent examination among various diagnostic nuclear medicine procedures. It is a well-established imaging modality for the diagnosis of osseous metastasis and for monitoring osseous tumor response to chemotherapy and radiation therapy. Although the sensitivity of bone scan examinations for detection of bone abnormalities has been considered to be relatively high, it is time consuming to identify multiple lesions such as bone metastases of prostate and breast cancers. In addition, it is very difficult to detect subtle interval changes between two successive abnormal bone scans, because of variations in patient conditions, the accumulation of radioisotopes during each examination, and the image quality of gamma cameras. Therefore, we developed a new computer-aided diagnostic (CAD) scheme for the detection of interval changes in successive whole-body bone scans by use of a temporal subtraction image which was obtained with a nonlinear image-warping technique. We carried out 58 pairs of successive bone scans in which each scan included both posterior and anterior views. We determined 107 "gold-standard" interval changes among the 58 pairs based on the consensus of three radiologists. Our computerized scheme consisted of seven steps, i.e., initial image density normalization on each image, image matching for the paired images, temporal subtraction by use of the nonlinear image-warping technique, initial detection of interval changes by use of temporal-subtraction images, image feature extraction of candidates of interval changes, rule-based tests by use of 16 image features for removing some false positives, and display of the computer output for identified interval changes. One hundred seven gold standard interval changes included 71 hot lesions (uptake was increased compared with the previous scan, or there was new uptake in the current scan) and 36 cold lesions (uptake was decreased or disappeared) for anterior and posterior views. The overall sensitivity in the detection of interval changes, including both hot and cold lesions evaluated by use of the resubstitution and the leave-one-case-out methods, were 95.3%, with 5.97 false positives per view, and 83.2% with 6.02, respectively. The temporal subtraction image for successive whole-body bone scans has the potential to enhance the interval changes between two images, which also can be quantified. Furthermore, the CAD scheme for the detection of interval changes by use of temporal subtraction images would be useful in assisting radiologists' interpretation on successive bone scan images.

A fully automated method for lung nodule detection from postero-anterior chest radiographs

In the past decades, a great deal of research work has been devoted to the development of systems that could improve radiologists' accuracy in detecting lung nodules. Despite the great efforts, the problem is still open. In this paper, we present a fully automated system processing digital postero-anterior (PA) chest radiographs, that starts by producing an accurate segmentation of the lung field area. The segmented lung area includes even those parts of the lungs hidden behind the heart, the spine, and the diaphragm, which are usually excluded from the methods presented in the literature. This decision is motivated by the fact that lung nodules may be found also in these areas. The segmented area is processed with a simple multiscale method that enhances the visibility of the nodules, and an extraction scheme is then applied to select potential nodules. To reduce the high number of false positives extracted, cost-sensitive support vector machines (SVMs) are trained to recognize the true nodules. Different learning experiments were performed on two different data sets, created by means of feature selection, and employing Gaussian and polynomial SVMs trained with different parameters; the results are reported and compared. With the best SVM models, we obtain about 1.5 false positives per image (fp/image) when sensitivity is approximately equal to 0.71; this number increases to about 2.5 and 4 fp/image when sensitivity is = 0.78 and = 0.85, respectively. For the highest sensitivity (= 0.92 and 1.0), we get 7 or 8 fp/image.

Feature subset selection for improving the performance of false positive reduction in lung nodule CAD

We propose a feature subset selection method based on genetic algorithms to improve the performance of false positive reduction in lung nodule computer-aided detection (CAD). It is coupled with a classifier based on support vector machines. The proposed approach determines automatically the optimal size of the feature set, and chooses the most relevant features from a feature pool. Its performance was tested using a lung nodule database (52 true nodules and 443 false ones) acquired by multislice CT scans. From 23 features calculated for each detected structure, the suggested method determined ten to be the optimal feature subset size, and selected the most relevant ten features. A support vector machine classifier trained with the optimal feature subset resulted in 100% sensitivity and 56.4% specificity using an independent validation set. Experiments show significant improvement achieved by a system incorporating the proposed method over a system without it. This approach can be also applied to other machine learning problems; e.g. computer-aided diagnosis of lung nodules.

Computer-aided diagnosis for the detection and classification of lung cancers on chest radiographs ROC analysis of radiologists' performance

RATIONALE AND OBJECTIVES

The aim of the study is to investigate the effect of a computer-aided diagnostic (CAD) scheme on radiologist performance in the detection of lung cancers on chest radiographs.

MATERIALS AND METHODS

We combined two independent CAD schemes for the detection and classification of lung nodules into one new CAD scheme by use of a database of 150 chest images, including 108 cases with solitary pulmonary nodules and 42 cases without nodules. For the observer study, we selected 48 chest images, including 24 lung cancers, 12 benign nodules, and 12 cases without nodules, from the database to investigate radiologist performance in the detection of lung cancers. Nine radiologists participated in a receiver operating characteristic (ROC) study in which cases were interpreted first without and then with computer output, which indicated locations of possible lung nodules, together with a five-color scale illustrating the computer-estimated likelihood of malignancy of the detected nodules.

RESULTS

Performance of the CAD scheme indicated that sensitivity in detecting lung nodules was 80.6%, with 1.2 false-positive results per image, and sensitivity and specificity for classification of nodules by use of the same database for training and testing the CAD scheme were 87.7% and 66.7%, respectively. Average area under the ROC curve value for detection of lung cancers improved significantly (P = .008) from without (0.724) to with CAD (0.778).

CONCLUSION

This type of CAD scheme, which includes two functions, namely detection and classification, can improve radiologist accuracy in the diagnosis of lung cancer.

How can a massive training artificial neural network (MTANN) be trained with a small number of cases in the distinction between nodules and vessels in thoracic CT?

RATIONALE AND OBJECTIVES

To demonstrate that a massive training artificial neural network (MTANN) can be adequately trained with a small number of cases in the distinction between nodules and vessels (non-nodules) in thoracic computed tomography (CT) images.

MATERIALS AND METHODS

An MTANN is a trainable, highly nonlinear filter consisting of a linear-output multilayer artificial neural network model. For enhancement of nodules and suppression of vessels, we used 10 nodules and 10 non-nodule images as training cases for MTANNs. The MTANN is trained with a large number of input subregions selected from the training cases and the corresponding pixels in teaching images that contain Gaussian distributions for nodules and zero for non-nodules. We trained three MTANNs with different numbers (1, 9, and 361) of training samples (pairs of the subregion and the teaching pixel) selected from the training cases. In order to investigate the basic characteristics of the trained MTANNs, we applied the MTANNs to simulated CT images containing various-sized model nodules (spheres) with different contrasts and various-sized model vessels (cylinders) with different orientations. In addition, we applied the trained MTANNs to nontraining actual clinical cases with 59 nodules and 1,726 non-nodules.

RESULTS

In the output images for the simulated CT images by use of the MTANNs trained with small numbers (one and nine) of subregions, model vessels were clearly visible and were not removed; thus, the MTANNs were not trained properly. However, in the output image of the MTANN trained with a large number of subregions, various-sized model nodules with different contrasts were represented by light nodular distributions, whereas various-sized model vessels with different orientations were dark and thus were almost removed. This result indicates that the MTANN was able to learn, from a very small number of actual nodule and non-nodule cases, the distinction between nodules (spherelike objects) and vessels (cylinder-like objects). In nontraining clinical cases, the MTANN was able to distinguish actual nodules from actual vessels in CT images. For 59 actual nodules and 1,726 non-nodules, the performance of the MTANN decreased as the number of training samples (subregions) in each case decreased.

CONCLUSIONS

The MTANN can be trained with a very small number of training cases (10 nodules and 10 non-nodules) in the distinction between nodules and non-nodules (vessels) in CT images. Massive training by scanning of training cases to produce a large number of training samples (input subregions and teaching pixels) would contributed to a high generalization ability of the MTANN.

False-positive reduction in computer-aided diagnostic scheme for detecting nodules in chest radiographs by means of massive training artificial neural network

RATIONALE AND OBJECTIVE

We developed a technique that uses a multiple massive-training artificial neural network (multi-MTANN) to reduce the number of false-positive results in a computer-aided diagnostic (CAD) scheme for detecting nodules in chest radiographs.

MATERIALS AND METHODS

Our database consisted of 91 solitary pulmonary nodules, including 64 malignant nodules and 27 benign nodules, in 91 chest radiographs. With our current CAD scheme based on a difference-image technique and linear discriminant analysis, we achieved a sensitivity of 82.4%, with 4.5 false positives per image. We developed the multi-MTANN for further reduction of the false positive rate. An MTANN is a highly nonlinear filter that can be trained with input images and corresponding teaching images. To reduce the effects of background levels in chest radiographs, we applied a background-trend-correction technique, followed by contrast normalization, to the input images for the MTANN. For enhancement of nodules, the teaching image was designed to contain the distribution for a "likelihood of being a nodule." Six MTANNs in the multi-MTANN were trained by using typical nodules and six different types of non-nodules (false positives).

RESULTS

Use of the trained multi-MTANN eliminated 68.3% of false-positive findings with a reduction of one true-positive result. The false-positive rate of our original CAD scheme was improved from 4.5 to 1.4 false positives per image, at an overall sensitivity of 81.3%.

CONCLUSION

Use of a multi-MTANN substantially reduced the false-positive rate of our CAD scheme for lung nodule detection on chest radiographs, while maintaining a level of sensitivity.

Wavelet-packet-based texture analysis for differentiation between benign and malignant liver tumours in ultrasound images

The purpose of this study was to apply a novel method of multiscale echo texture analysis for distinguishing benign (hemangiomas) from malignant (hepatocellular carcinomas (HCCs) and metastases) focal liver lesions in B-mode ultrasound images. In this method, regions of interest (ROIs) extracted from within the lesions were decomposed into subimages by wavelet packets. Multiscale texture features that quantify homogeneity of the echogenicity were calculated from these subimages and were combined by an artificial neural network (ANN). A subset of the multiscale features was selected that yielded the highest performance in the classification of lesions measured by the area under the receiver operating characteristic curve (Az). In an analysis of 193 ROIs consisting of 50 hemangiomas, 87 hepatocellular carcinomas and 56 metastases, the multiscale features yielded a high A: value of 0.92 in distinguishing benign from malignant lesions, 0.93 in distinguishing hemangiomas from HCCs and 0.94 in distinguishing hemangiomas from metastases. Our new multiscale texture analysis method can effectively differentiate malignant from benign lesions, and thus has the potential to increase the accuracy of diagnosis of focal liver lesions in ultrasound images.

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.

Prediction of survival in surgical unresectable lung cancer by artificial neural networks including genetic polymorphisms and clinical parameters

Lung cancer, a common malignancy in Taiwan, involves multiple factors, including genetics and environmental factors. The survival time is very short once cancer is diagnosed as being in advanced stage and surgically unresectable. Therefore, a good model of prediction of disease outcome is important for a treatment plan. We investigated the survival time in advanced lung cancer by using computer science from the genetic polymorphism of the p21 and p53 genes in conjunction with patients' general data. We studied 75 advanced and surgical unresectable lung cancer patients. The prediction of survival time was made by comparing real data obtained from follow-up periods with data generated by an artificial neural network (ANN). The most important input variable was the clinical staging of lung cancer patients. The second and third most important variables were pathological type and responsiveness to treatment, respectively. There were 25 neurons in the input layer, four neurons in the hidden layer-1, and one neuron in the output layer. The predicted accuracy was 86.2%. The average survival time was 12.44 +/- 7.95 months according to real data and 13.16 +/- 1.77 months based on the ANN results. ANN provides good prediction results when clinical parameters and genetic polymorphisms are considered in the model. It is possible to use computer science to integrate the genetic polymorphisms and clinical parameters in the prediction of disease outcome. Data mining provides a promising approach to the study of genetic markers for advanced lung cancer.

Computerized classification of suspicious regions in chest radiographs using subregion Hotelling observers

We propose to investigate the use of subregion Hotelling observers (SRHOs) in conjunction with perceptrons for the computerized classification of suspicious regions in chest radiographs for being nodules requiring follow up. Previously, 239 regions of interest (ROIs), each containing a suspicious lesion with proven classification, were collected. We chose to investigate the use of SRHOs as part of a multilayer classifier to determine the presence of a nodule. Each SRHO incorporates information about signal, background, and noise correlation for classification. For this study, 225 separate Hotelling observers were set up in a grid across each ROI. Each separate observer discriminates an 8 by 8 pixel area. A round robin sampling scheme was used to generate the 225 features, where each feature is the output of the individual observers. These features were then rank ordered by the magnitude of the weights of a perceptron. Once rank ordered, subsets of increasing number of features were selected to be used in another perceptron. This perceptron was trained to minimize mean squared error and the output was a continuous variable representing the likelihood of the region being a nodule. Performance was evaluated by receiver operating characteristic (ROC) analysis and reported as the area under the curve (Az). The classifier was optimized by adding additional features until the Az declined. The optimized subset of observers then were combined using a third perceptron. A subset of 80 features was selected which gave an Az of 0.972. Additionally, at 98.6% sensitivity, the classifier had a specificity of 71.3% and increased the positive predictive value from 60.7% to 84.1 %. Preliminary results suggest that using SRHOs in combination with perceptrons can provide a successful classification scheme for pulmonary nodules. This approach could be incorporated into a larger computer aided detection system for decreasing false positives.

Computer-aided diagnosis in chest radiography: a survey

The traditional chest radiograph is still ubiquitous in clinical practice, and will likely remain so for quite some time. Yet, its interpretation is notoriously difficult. This explains the continued interest in computer-aided diagnosis for chest radiography. The purpose of this survey is to categorize and briefly review the literature on computer analysis of chest images, which comprises over 150 papers published in the last 30 years. Remaining challenges are indicated and some directions for future research are given.

Computer-aided diagnostic scheme for lung nodule detection in digital chest radiographs by use of a multiple-template matching technique

We have been developing a computer-aided diagnostic (CAD) scheme to assist radiologists in improving the detection of pulmonary nodules in chest radiographs, because radiologists can miss as many as 30% of pulmonary nodules in routine clinical practice. A key to the successful clinical application of a CAD scheme is to ensure that there are only a small number of false positives that are incorrectly reported as nodules by the scheme. In order to significantly reduce the number of false positives in our CAD scheme, we developed, in this study, a multiple-template matching technique, in which a test candidate can be identified as a false positive and thus eliminated, if its largest cross-correlation value with non-nodule templates is larger than that with nodule templates. We describe the technique for determination of cross-correlation values for test candidates with nodule templates and non-nodule templates, the technique for creation of a large number of nodule templates and non-nodule templates, and the technique for removal of nodulelike non-nodule templates and non-nodulelike nodule templates, in order to achieve a good performance. In our study, a large number of false positives (44.3%) were removed with reduction of a very small number of true positives (2.3%) by use of the multiple-template matching technique. We believe that this technique can be used to significantly improve the performance of CAD schemes for lung nodule detection in chest radiographs.

Automatic detection and quantification of ground-glass opacities on high-resolution CT using multiple neural networks: comparison with a density mask

OBJECTIVE

We compared multiple neural networks with a density mask for the automatic detection and quantification of ground-glass opacities on high-resolution CT under clinical conditions.

SUBJECTS AND METHODS

Eighty-four patients (54 men and 30 women; age range, 18-82 years; mean age, 49 years) with a total of 99 consecutive high-resolution CT scans were enrolled in the study. The neural network was designed to detect ground-glass opacities with high sensitivity and to omit air-tissue interfaces to increase specificity. The results of the neural network were compared with those of a density mask (thresholds, -750/-300 H), with a radiologist serving as the gold standard.

RESULTS

The neural network classified 6% of the total lung area as ground-glass opacities. The density mask failed to detect 1.3%, and this percentage represented the increase in sensitivity that was achieved by the neural network. The density mask identified another 17.3% of the total lung area to be ground-glass opacities that were not detected by the neural network. This area represented the increase in specificity achieved by the neural network. Related to the extent of the ground-glass opacities as classified by the radiologist, the neural network (density mask) reached a sensitivity of 99% (89%), specificity of 83% (55%), positive predictive value of 78% (18%), negative predictive value of 99% (98%), and accuracy of 89% (58%).

CONCLUSION

Automatic segmentation and quantification of ground-glass opacities on high-resolution CT by a neural network are sufficiently accurate to be implemented for the preinterpretation of images in a clinical environment; it is superior to a double-threshold density mask.

On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology

The application of artificial neural networks (ANNs) for prognostic and diagnostic classification in clinical medicine has become very popular. In particular, feed-forward neural networks have been used extensively, often accompanied by exaggerated statements of their potential. In this paper, the essentials of feed-forward neural networks and their statistical counterparts (that is, logistic regression models) are reviewed. We point out that the uncritical use of ANNs may lead to serious problems, such as the fitting of implausible functions to describe the probability of class membership and the underestimation of misclassification probabilities. In applications of ANNs to survival data, further difficulties arise. Finally, the results of a search in the medical literature from 1991 to 1995 on applications of ANNs in oncology and some important common mistakes are reported. It is concluded that there is no evidence so far that application of ANNs represents real progress in the field of diagnosis and prognosis in oncology.

Characteristics of regions suspicious for pulmonary nodules at chest radiography

RATIONALE AND OBJECTIVES

This study was performed to determine physical characteristics of areas on chest radiographs that are suspicious but not definitive for the presence of a pulmonary nodule and the characteristics of areas that contain an obvious nodule.

MATERIALS AND METHODS

Two groups of patients were identified: those who had an area at plain radiography that was suspicious for a pulmonary nodule and underwent fluoroscopy for further evaluation (138 patients, 142 areas) and those who had an obvious nodule at plain radiography who underwent computed tomography for further evaluation (72 patients, 97 areas). The measured characteristics of the region of interest included size, circularity, compactness, contrast, and location.

RESULTS

A comparison of the data show that while there was some difference between these groups of patients with regard to location of the nodules, there were essentially no differences with regard to size, circularity, compactness, and contrast of the regions of interest.

CONCLUSION

Size, circularity, compactness, contrast, and location are not sufficient to distinguish pulmonary nodules from other suspicious regions on the chest radiograph.

Fragmentary window filtering for multiscale lung nodule detection: preliminary study

RATIONALE AND OBJECTIVES

The authors evaluated computer-assisted diagnostic (CAD) methods used to detect suspicious areas on lung radiographs.

MATERIALS AND METHODS

The authors designed a fragmentary window filtering (FWF) algorithm for detecting lung nodule patterns, which generally appear as circular areas of high opacity on the chest radiograph. The FWF algorithm helps differentiate circular patterns from overlapping radiographic background. A multiscale analysis was performed to locate multiscale nodules. Receiver operating characteristic analysis was performed by using a lung nodule that was extracted from a chest radiograph. The nodule underwent scalings and subsequent superimposition onto 140 normal regions of interest from six chest radiographs.

RESULTS

The FWF method was superior to the matched filtering method in the detection of suspicious areas.

CONCLUSION

The proposed FWF-based method should provide improved detection of lung nodules on chest radiographs.

Fuzzy logic in computer-aided breast cancer diagnosis: analysis of lobulation

This paper illustrates how a fuzzy logic approach can be used to formalize terms in the American College of Radiology (ACR) Breast Imaging Lexicon. In current practice, radiologists make a relatively subjective determination for many terms from the lexicon related to breast cancer diagnosis. Lobulation and microlobulation of nodules are two important features in the ACR lexicon. We offer an approach for formalizing the distinction of these features and also formalize the description of intermediate cases between lobulated and microlobulated masses. In this paper it is shown that fuzzy logic can be an effective tool in dealing with this kind of problem. The proposed formalization creates a basis for the next three steps (i) extended verification with blinded comparison studies. (ii) the automatic extraction of the related primitives from the image, and (iii) the detection of lobulated and microlobulated masses based on these primitives.

Fractal texture analysis in computer-aided diagnosis of solitary pulmonary nodules

RATIONALE AND OBJECTIVES

The authors investigated the use of fractal texture characterization to improve the accuracy of solitary pulmonary nodule computer-aided diagnosis (CAD) systems.

METHODS

Thirty chest radiographs were acquired from patients who had no pulmonary nodules. Thirty regions were selected that were considered remotely suspicious-looking for nodules. Artificial nodules of multiple shapes, sizes, and orientations were added at subtle levels of contrast to 30 non-suspicious-looking regions of the radiographs. Fractal dimensions of the 60 "nodule candidates" were calculated to quantify the texture of each region. Four radiologists also interpreted the images.

RESULTS

The fractal dimension of each possible nodule provided statistically significant (P < .05) differentiation between regions that contained an artificial nodule and those that did not. The area under the receiver operating characteristic curve for the fractal analysis was significantly better (P < .05) than that for the radiologists.

CONCLUSION

Fractal texture characterization provides useful information for the classification of potential solitary pulmonary nodules with CAD algorithms.

Reduction of false positives in lung nodule detection using a two-level neural classification

The authors have developed a neural-digital computer-aided diagnosis system, based on a parameterized two-level convolution neural network (CNN) architecture and on a special multilabel output encoding procedure. The developed architecture was trained, tested, and evaluated specifically on the problem of diagnosis of lung cancer nodules found on digitized chest radiographs. The system performs automatic "suspect" localization, feature extraction, and diagnosis of a particular pattern-class aimed at a high degree of "true-positive fraction" detection and low "false-positive fraction" detection. In this paper, the authors aim at the presentation of the two-level neural classification method in reducing false-positives in their system. They employed receiver operating characteristics (ROC) method with the area under the ROC curve (A(z)) as the performance index to evaluate all the simulation results. The two-level CNN showed superior performance (A(z)=0.93) to the single-level CNN (A(z)=0.85). The proposed two-level CNN architecture is proven to be promising and to be extensible, problem-independent, and therefore, applicable to other medical or difficult diagnostic tasks in two-dimensional (2-D) image environments.

Detection of lung nodules in digital chest radiographs using artificial neural networks: a pilot study

Radiologists can fail to detect up to 30% of pulmonary nodules in chest radiographs. A back-propagation neural network was used to detect lung nodules in digital chest radiographs to assist radiologists in the diagnosis of lung cancer. Regions of interest (ROIs) that contained nodules and normal tissues in the lung were selected from digitized chest radiographs by a previously developed computer-aided diagnosis (CAD) scheme. Different preprocessing techniques were used to produce input data to the neural network. The performance of the neural network was evaluated by receiver operating characteristic (ROC) analysis. We found that subsampling of original 64- x 64-pixel ROIs to smaller 8- x 8-pixel ROIs provides the optimal preprocessing for the neural network to distinguish ROIs containing nodules from false-positive ROIs containing normal regions. The neural network was able to detect obvious nodules very well with an Az value (area under ROC curve) of 0.93, but was unable to detect subtle nodules. However, with a training method that uses different orientations of the original ROIs, we were able to improve the performance of the neural network to detect subtle nodules. Artificial neural networks have the potential to serve as a useful classifier to help to eliminate the false-positive detections of the CAD scheme.