Effect of Texture Features in Computer Aided Diagnosis of Pulmonary Nodules in Low-Dose Computed Tomography

Krewer, Henry; Geiger, Benjamin; Hall, Lawrence O.; Goldgof, Dmitry B.; Gu, Yi; Tockman, Melvyn S.; Gillies, Robert J.

doi:10.1109/smc.2013.663

Cited by 38 publications

(19 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Statistics from the matrix were computed and used as inputs to a linear-discriminant classifier for classifying 96 nodules either as malignant or benign, achieving an area under the receiver operating characteristic (ROC) curve [area-under-the-curve (AUC)] score of 0.83. Krewer et al 13 automatically extracted 219 2-D and 3-D texture and shape features and used a feature-selection method to find significant features, which were then used as inputs to support vector machines (SVMs), decision trees, and nearest-neighbor classification methods on 33 cases from the LIDC dataset. They reported an accuracy of 90.91% in classifying 14 malignant and 19 benign nodules from the LIDC dataset when correlation-based feature-selection was used.…”

Section: Previous Work Related To Approach Onementioning

confidence: 99%

“…[5][6][7][8][9][10][11][12] On the other hand, others have used the algorithmic quantification of image features only as intermediate quantities within a system for classifying nodules as malignant or benign. [13][14][15][16][17][18] In these approaches, it is not clear how well the quantified features capture the true physical nature of the features themselves because only error metrics for the accuracy of nodule classification are considered rather than the approximation error in quantifying the features themselves.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Lung nodule malignancy classification using only radiologist-quantified image features as inputs to statistical learning algorithms: probing the Lung Image Database Consortium dataset with two statistical learning methods

Hancock

Magnan

2016

J. Med. Imag

View full text Add to dashboard Cite

In the assessment of nodules in CT scans of the lungs, a number of image-derived features are diagnostically relevant. Currently, many of these features are defined only qualitatively, so they are difficult to quantify from first principles. Nevertheless, these features (through their qualitative definitions and interpretations thereof) are often quantified via a variety of mathematical methods for the purpose of computer-aided diagnosis (CAD). To determine the potential usefulness of quantified diagnostic image features as inputs to a CAD system, we investigate the predictive capability of statistical learning methods for classifying nodule malignancy. We utilize the Lung Image Database Consortium dataset and only employ the radiologist-assigned diagnostic feature values for the lung nodules therein, as well as our derived estimates of the diameter and volume of the nodules from the radiologists' annotations. We calculate theoretical upper bounds on the classification accuracy that are achievable by an ideal classifier that only uses the radiologist-assigned feature values, and we obtain an accuracy of 85.74 [Formula: see text], which is, on average, 4.43% below the theoretical maximum of 90.17%. The corresponding area-under-the-curve (AUC) score is 0.932 ([Formula: see text]), which increases to 0.949 ([Formula: see text]) when diameter and volume features are included and has an accuracy of 88.08 [Formula: see text]. Our results are comparable to those in the literature that use algorithmically derived image-based features, which supports our hypothesis that lung nodules can be classified as malignant or benign using only quantified, diagnostic image features, and indicates the competitiveness of this approach. We also analyze how the classification accuracy depends on specific features and feature subsets, and we rank the features according to their predictive power, statistically demonstrating the top four to be spiculation, lobulation, subtlety, and calcification.

show abstract

Section: Previous Work Related To Approach Onementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Lung nodule malignancy classification using only radiologist-quantified image features as inputs to statistical learning algorithms: probing the Lung Image Database Consortium dataset with two statistical learning methods

Hancock

Magnan

2016

J. Med. Imag

View full text Add to dashboard Cite

show abstract

“…Among them, feature selection and classification are key steps to improve the sensitivity and accuracy of the entire system. The conventional CAD systems [3][4][5][6] are fundamentally based on complex pattern recognition, which highly rely on image processing to capture reliable features. Despite the favorable performance of these CAD systems in lung nodules analysis, the extracted features tend to be subjective, which limits the performance of the model to a certain extent.…”

Section: Introductionmentioning

confidence: 99%

Synthetic Medical Images Using F&BGAN for Improved Lung Nodules Classification by Multi-Scale VGG16

Zhao

Zhu

et al. 2018

Symmetry

View full text Add to dashboard Cite

Lung cancer is one of the highest causes of cancer-related death in both men and women. Therefore, various diagnostic methods for lung nodules classification have been proposed to implement the early detection. Due to the limited amount and diversity of samples, these methods encounter some bottlenecks. In this paper, we intend to develop a method to enlarge the dataset and enhance the performance of pulmonary nodules classification. We propose a data augmentation method based on generative adversarial network (GAN), called Forward and Backward GAN (F&BGAN), which can generate high-quality synthetic medical images. F&BGAN has two stages, Forward GAN (FGAN) generates diverse images, and Backward GAN (BGAN) is used to improve the quality of images. Besides, a hierarchical learning framework, multi-scale VGG16 (M-VGG16) network, is proposed to extract discriminative features from alternating stacked layers. The methodology was evaluated on the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset, with the best accuracy of 95.24%, sensitivity of 98.67%, specificity of 92.47% and area under ROC curve (AUROC) of 0.980. Experimental results demonstrate the feasibility of F&BGAN in generating medical images and the effectiveness of M-VGG16 in classifying malignant and benign nodules.

show abstract

“…Methods An electronic learning module was developed by the author based on existing NICE guidelines. 2 Candidates described a case of a smoker they had seen and were quizzed about the case by the learning module with some instant feedback, then deeper discussion with the author via e-mail. When the candidate was ready a CBD form was completed.…”

Section: The Smoking Gunmentioning

confidence: 99%

S77 Processing of Cigarette Graphic Health Warning Labels Decrease with Prolonged Exposure: Abstract S77 Table 1

Ratneswaran

Chisnall

et al. 2015

Thorax

View full text Add to dashboard Cite

IntroductionCigarette package graphic health warning labels (GHWL) remain an important means of communicating serious smoking risks. There were significant delays in implementing them in different countries around the world, partly due to tobacco company resistance; additionally, the messages used differ considerably. We expected a reduced cognitive processing of the messages based on the duration since launch. In order to address this question we compared a London (4 years) vs Singapore (8 years exposed) cohort.MethodsWe used a 50-item structured interview; after recording demographics and smoking history, 10 country-specific warning labels were shown. In addition to the emotional response and impact on smoking cessation/prevention, cognitive processing was assessed on a scale from ‘1’ (‘not at all/never’) to ‘5’ (‘all the time/a lot’). Smoking-risk knowledge and their importance in terms of prevention and treatment were elicited.Results266 participants were recruited, 163 from London (52 ± 18 years, 54% male, 35% smokers) and 103 from Singapore (58 ± 15 years, p = 0.012; 78% male, p < 0.001; 53% smokers, p = 0.003). Londoners read the labels more carefully and more often; they talked and thought more about them, even with no warning labels were in sight, and they kept packages more often as a reminder about their messages (overall, 2.0 ± 1.3 vs 1.5 ± 1.0, p < 0.001). The processing differences between the cities were consistent when comparing the London and Singapore smokers (overall, 2.0 ± 1.1 vs 1.6 ± 1.1, p < 0.001) and non-smokers (overall, 2.0 ± 1.4 vs 1.4 ± 0.8, p < 0.001) (Table 1). Londoners experienced more disgust when viewing the images (79% vs 53%, p < 0.001) and felt they were more effective deterrents (51% vs 35%, p = 0.011). One-in-five participants in Singapore were unaware of the association between smoking and lung cancer, despite it being the most deterring risk; blindness was the least well-known consequence in London (24%) and Singapore (34%, p = 0.075) despite being ranked ahead of stroke, oral cancer, and in smokers, ahead of heart disease for importance to prevent/treat.Abstract S77 Table 1Processing of GHWL; comparing all participants, non-smokers and smokers in London vs SingaporeAllNon-smokersSmokersProcessing (/5) (SD)SingLondonpSingLondonpSingLondonpCarefully read labels1.8(1.1)2.2(1.2)0.0041.6(0.9)2.1(1.3)0.0241.9(1.2)2.4(1.1)0.028Often read labels1.7(1.1)2.2(1.3) <0.0011.5(0.9)2.2(1.4)0.0011.9(1.2)2.3(1.0)0.095Ever talked about1.5(1.1)2.2(1.4) <0.0011.6(1.1)2.3(1.5)0.0011.5(1.2)2.0(1.1)0.039Often think about1.6(1.0)2.4(1.4) <0.0011.5(0.9)2.5(1.4)0.0011.8(1.1)2.4(1.3)0.006Inc. when not in sight1.3(0.7)1.8(1.1) <0.0011.1(0.6)1.8(1.2)0.0011.4(0.9)1.8(1.0)0.040Kept labels as reminder1.0(0.3)1.3(0.8)0.0011.0(0.2)1.3(0.9)0.0211.0(0.3)1.3(0.8)0.023ConclusionA desensitisation to graphic health warning labels occurs with extended exposure. Non-smokers are prone to the same desensitisation as smokers are. In pre-empting this, the awareness and impact of specific health risks need to be activ...

show abstract

Effect of Texture Features in Computer Aided Diagnosis of Pulmonary Nodules in Low-Dose Computed Tomography

Cited by 38 publications

References 10 publications

Lung nodule malignancy classification using only radiologist-quantified image features as inputs to statistical learning algorithms: probing the Lung Image Database Consortium dataset with two statistical learning methods

Lung nodule malignancy classification using only radiologist-quantified image features as inputs to statistical learning algorithms: probing the Lung Image Database Consortium dataset with two statistical learning methods

Synthetic Medical Images Using F&BGAN for Improved Lung Nodules Classification by Multi-Scale VGG16

S77 Processing of Cigarette Graphic Health Warning Labels Decrease with Prolonged Exposure: Abstract S77 Table 1

Contact Info

Product

Resources

About