Research Presentation Session: Chest
01:17Lorenzo E. Derchi, Leyla Musayeva
07:05Kiran Vaidhya Venkadesh
Author Block: K. V. Venkadesh, T. A. Aleef, A. Schreuder, E. Scholten, B. Van Ginneken, M. Prokop, C. Jacobs; Nijmegen/NL
Purpose or Learning Objective: Nodule size, morphology, and growth are important factors for accurately estimating nodule malignancy risk in lung cancer screening CT examinations. In this work, we aimed to develop a deep learning (DL) algorithm that uses a current and a prior CT examination to estimate the malignancy risk of pulmonary nodules.
Methods or Background: We developed a dual time-point DL algorithm by stacking the nodules from the current and prior CT examinations in the input channels of convolutional neural networks. We used 3,011 nodules (286 malignant) and 994 nodules (73 malignant) as development and hold-out test cohorts from the National Lung Screening Trial, respectively. The reference standard was set by histopathologic confirmation or CT follow-up of more than two years. We compared the performance of the algorithm against PanCan model 2b and a previously published single time-point DL algorithm that only processed a single CT examination. We used the area under the receiver operating characteristic curve (AUC) to measure discrimination performance and a standard permutation test with 10,000 random permutations to compute p-values.
Results or Findings: The dual time-point DL algorithm achieved an AUC of 0.94 (95% CI: 0.91 - 0.97) on the hold-out test cohort. The algorithm outperformed the single time-point DL algorithm and the PanCan model, which had AUCs of 0.92 (95% CI: 0.89 - 0.95; p = 0.055) and 0.88 (95% CI: 0.85 - 0.91; p < 0.001), respectively.
Conclusion: Deep learning algorithms using current and prior CT examinations have the potential to accurately estimate the malignancy risk of pulmonary nodules.
Limitations: External validation is needed on other screening datasets to generate further evidence.
Ethics committee approval: Institutional review board approval was obtained at each of the 33 centers involved in the NLST.
Funding for this study: Research grant from MeVis Medical Solutions AG.
06:39Di Tian
Author Block: D. Tian, Y. He, J. Zhang, Q. Song, A. Liu, Z. Li; Dalian/CN
Purpose or Learning Objective: Radiomics for classifying minimally invasive lung adenocarcinoma and invasive lung adenocarcinoma presenting as ground-glass nodules based on contrast-enhanced computed tomography.
Methods or Background: We retrospectively included 98 ground glass nodules (GGNs) that were surgically confirmed as minimally invasive adenocarcinomas (MIAs) or invasive adenocarcinomas (IAs). Each GGO was segmented manually using 3D-slicer, and texture features were extracted from the enhanced venous phase image. The least absolute shrinkage and selection operator (LASSO) method was applied to select optimal radiomics features whose performance was assessed by the area under the receiver operating characteristic curve (AUC-ROC). The radiomics model was compared to the radiographic model and the radiomics-radiographic CT combined model using univariate and multivariate logistic regression analysis.
Results or Findings: The radiomics model in CT-enhanced venous phase showed better discriminative performance (training AUC,0.85; test AUC,0.85) than the radiographic CT model (training AUC,0.78; test AUC,0.62). The combined model (training AUC,0.85; test AUC,0.85) did not demonstrated improved performance compared with the radiomics model.
Conclusion: A radiomics model based on contrast-enhanced CT imaging have the best diagnostic performance to distinguish MIAs from IAs in GGNs when compare with the radiographic CT model.
Limitations: It was a single-center retrospective study, and there is deviation between collection and inclusion.
Ethics committee approval: The ethics committee of the First Affiliated Hospital of Dalian Medical University.
Funding for this study: No funding support.
05:14Jakob Leonhardi
Author Block: J. Leonhardi, J. Pappisch, A-K. Höhn, H. Wirtz, T. Denecke, A. Frille, H-J. Meyer; Leipzig/DE
Purpose or Learning Objective: Texture analysis derived from computed tomography (CT) might be able to provide clinically relevant imaging biomarkers and might be associated with histopathology features in tumours. The present study sought to elucidate possible associations between texture features derived from CT images with proliferation index Ki-67 and grading in pulmonary neuroendocrine tumours.
Methods or Background: 38 patients (n= 22 females, 58%) with a mean age of 60.8 ± 15.2 years were included into this retrospective study. Texture analysis was performed using the free available Mazda software. All tumours were histopathologically confirmed. Discrimination and correlation analyses were performed.
Results or Findings: In discrimination analysis, "S(1,1)SumEntrp" was significantly different between typical and atypical carcinoids (mean 1.74 ± 0.11 versus 1.79 ± 0.14, p=0.007). The correlation analysis revealed a moderate positive association between Ki-67 index with the first order parameter kurtosis (r=0.66, p=0.001). Several other texture features were associated with Ki-67 index, the highest correlation coefficient showed S(4,4)InvDfMom" (r=0.59, p=0.004).
Conclusion: Several texture features derived from CT were associated with proliferation index Ki-67 and might therefore be a valuable novel biomarker in pulmonary neuroendocrine tumours. "Sumentrp" might be a promising parameter to aid in the discrimination between typical and atypical carcinoids.
Limitations: First, it is a retrospective study with possible known inherent bias. Second, the patient sample is rather small based upon pNET low prevalence among lung tumours. Third, only 60% of patients had a Ki-67 index available, which further reduces the size of the patient sample.
Ethics committee approval: It received ethical approval from the local ethics committee at the Medical Faculty (IRB00001750, AZ: 259/18-ek) on July 31 2018.
Funding for this study: No funding was provided for this study.
05:00Caro Franck
Author Block: C. Franck1, F. Zanca2, K. Carpentier1, H. El Addouli1, M. Spinhoven1, M. C. Niekel1, A. Van Hoyweghen1, A. Snoeckx1; 1Edegem/BE, 2Heverlee/BE
Purpose or Learning Objective: Deep learning image reconstruction (DLIR) has been shown to reduce radiation dose compared to iterative reconstruction (IR) for chest CT. It is unknown, however, whether the nodule volume measurements with DLIR are comparable with IR at different dose levels.
Methods or Background: An anthropomorphic chest phantom (Lungman, Kyoto Kagaku), containing six spherical, six lobulated, and six spiculated 3D-printed solid nodules (volume range 28-392 mm³), was scanned at six dose levels (0.2, 0.4, 0.8, 1.5, 3, 6 mGy). Images were 1.25 mm reconstructed with ASIR-V 60% and three levels of DLIR (TrueFidelity Low, Medium, High). The volumes of 432 nodules (18 nodules x 6 doses x 4 reconstructions) were measured by five independent observers in a semi-automatic fashion. Mean percentage error in nodule volume measurements was assessed for all reconstructions and dose levels, with respect to the ground truth (high dose scan, 11 mGy). Subsequently, data were stratified per nodule type. A smaller absolute percentage error indicates a higher accuracy.
Results or Findings: In general, mean % errors decreased with increasing dose. On average, errors were significantly lower with TrueFidelity (3.6/3.4/3.0% for Low/Medium/High) than with ASIR-V (4.1%), for all dose levels (p=0.001). With increasing DLIR level, errors decreased for the lower dose range (0.2-0.8 mGy), while for higher doses (1.5-6 mGy) values were comparable. When stratifying per morphology, the largest error was found for lobulated nodules (4.8%), followed by spiculated (3.3%) and spherical (2.8%) nodules.
Conclusion: In chest CT, volume measurements with TrueFidelity showed a significantly higher accuracy compared to ASIR-V, for all dose levels and all nodule types. Lobulated nodules showed the highest absolute error in volume measurements.
Limitations: Only phantom images were used.
Ethics committee approval: Not applicable.
Funding for this study: Not applicable.
06:34Jiansheng Qiu
Author Block: J. Qiu1, X. Chen2, X. Xin1, B. Zhang1; 1Nanjing/CN, 2Suzhou/CN
Purpose or Learning Objective: To investigate the clinical value of virtual monoenergetic images (VMI) and electron density map (EDM) derived from the dual-layer spectral detector CT (DLCT) in the differential diagnosis of benign and malignant pulmonary ground glass nodules (GGN).
Methods or Background: 27 benign and 38 malignant GGN were retrospectively studied. The 120kVp polyenergetic image (PI), EDM, and 40-80keV VMI were reconstructed, in which the CT value, electron density (ED), and CT features were analysed between benign and malignant lesions. The CT features included lesion size, location, shape, edge, internal structure, adjacent structure, and nodule type. The statistically significant CT signs and quantitative parameters were analysed by logistic regression analysis to obtain the independent risk factors for GGN malignancy, then all the independent risk factors were united to analyse by ROC curve.
Results or Findings: The lesion shape, spiculation, lobulation, location, size, CT value in PI, 40-80 keV VMI, and ED were significant different between two groups (P<0.05), which were enrolled in logistic regression analysis by taking if GGN was malignant as the dependent variable. The EDM had higher efficacy than other VMI with AUC 0.722. Logistic regression analysis results showed that ED (P=0.044), lesion size (P=0.004), and spiculation (P=0.002) were independent risk factors for GGN malignancy. ROC analysis showed that the AUC of ED, size, spiculation, and a combination of the three were 0.722, 0.772, 0.698, and 0.885, respectively. The AUC of combined diagnosis was the largest with sensitivity specificity of 92.1% and 74.1%, respectively.
Conclusion: The diagnostic efficiency of DLCT images in differential diagnosis of pulmonary GGN, EDM has a higher efficiency than other VMI, and the diagnostic efficiency is further improved when EDM combines with lesion size and spiculation is analysed comprehensively.
Limitations: Not applicable.
Ethics committee approval: Not applicable.
Funding for this study: Not applicable.
18:26Alexander Favril
Author Block: A. Favril, L. Vael, A. Snoeckx, C. Franck; Edegem/BE
Purpose or Learning Objective: Chest x-ray is frequently performed in daily clinical practice. The detection of lung nodules on radiographs is an indication for chest CT referral. Since lung nodule detection in chest x-ray has a low accuracy, the sensitivity can be improved by computer aided diagnosis (CAD) systems. However, the risk of large false positive rates can lead to redundant referrals for chest CT and contribute to unnecessary radiation exposure. In this retrospective study, we investigated the impact of an automatic lung nodule detection (ALND) algorithm on the evaluation of chest radiographs and referral for chest CT.
Methods or Background: Between July and September 2020, 1468 consecutive routine PA chest radiographs of adult, non-pregnant patients were retrospectively collected at a tertiary care center. Images were screened for positive ALND annotations in combination with referral for chest CT. In addition, the correlation of the ALND result with a true positive lung nodule was investigated.
Results or Findings: The ALND algorithm detected 591 lung nodules on 524 (36%) chest radiographs. Among these, 12 were indicated as known lesions, 301 were reported as false positive, whereas 220 ALND findings were not reported. On 9% (49/524) of the radiographs, 58 nodules were deemed suspicious and referred for CT. In 55% (27/49) of the cases a CT examination was actually performed of which in 56% (15/27) of the patients a nodular lesion was found.
Conclusion: In approximately 1 out of 10 patients, a positive annotation by an ALND algorithm leads to chest CT referral. When a subsequent chest CT is performed, the suspected nodule correlates with a true lung nodule in more than half of the investigated patients.
Limitations: Not applicable.
Ethics committee approval: Approved.
Funding for this study: Not applicable.