Citations

Abstract Rationale In fungal endemic regions, differentiating benign from malignant pulmonary nodules on CT scans remains a major diagnostic challenge. Coccidioides nodules frequently resemble malignant lesions on CT and PET imaging, presenting significant challenges for differentiation, even among radiologists. Consequently, the benign biopsy and resection rate, especially in the Southwestern US, remains high. We previously developed [DOI:10.1371/journal.pone.0196910] and externally validated [DOI:10.1183/13993003.02485-2020] the Benign versus aggRessive nODule Evaluation using Radiomic Stratification (BRODERS) machine learning (ML) classifier to non-invasively distinguish between benign and malignant pulmonary nodules with an AUC of 0.87. However, its performance in certain benign nodules with malignant surface and texture characteristics like in coccidioidomycosis had a suboptimal AUC of 0.65. In this study, we redesigned our radiomics model by including numerous clinical features. Methods In this retrospective study, 113 pulmonary nodules (66 Coccidioides and 47 adenocarcinoma spectrum) detected on CTs with ground truth pathology diagnosis were identified by screening for key terms like “nodule”, “cocci”, and “adenocarcinoma” from both CT and pathology reports using a large language model trained on clinical textual data (Illuminate®). Clinical features such as demographics, smoking history, and laboratory studies were extracted by trained clinicians from the electronic medical record. All nodules were segmented and radiomic features extracted using a semi-automated seed growing algorithm previously described [DOI: 10.1097/JTO.0b013e3182843721]. The database was randomly split into an 80%-20% development and hold-out set prior to ML modeling. Feature selection for variables was conducted utilizing Elastic Net Regression, which provides a balanced approach between Lasso and Ridge regression. A logistic regression model was developed using the selected features and ground truth information of the training cohort. The true performance of the model was tested on the hold-out set. Results 160 radiomics and 22 clinical features were extracted. Elastic Net regression identified 44 features that contributed the most. Some of these clinical features include presence of satellite lesions, smoking history, and presence of positive IgG coccidiomycosis serology. The AUC of the ROC curve for the hold-out test set was 0.86. Conclusions The clinical features of significance that emerged during multimodal modeling are explainable. Designing explainable ML models is important, especially in the Southwestern US where Coccidioides nodules can appear to be malignant to both human experts and ML models. Although coccidioidomycosis serologies alone lack sensitivity and specificity to diagnose Coccidioides nodules, integrating them with a multimodal ML model can improve differentiation between benign Coccidioides and malignant adenocarcinoma nodules. This abstract is funded by: None

Introduction Bronchoscopic lung volume reduction with endobronchial valves (EBV) is an effective procedure for patient with severe emphysema to improve lung function, exercise tolerance, dyspnoea and quality of life. Optimisation of patient and treatment lobe selection is essential for successful EBV outcomes. While clinical selection criteria are rigorous, many centres use a multidisciplinary team and rely on previous clinical experience for the selection process. To aid objective clinical decision making, we present a mathematical model to facilitate patient and target lobe selection. Methods A multidisciplinary team reviewed quantitative high-resolution computed tomography (HRCT) analysis from 119 patients to select candidates for EBV and to select a treatment lobe. Two logistic regression models, (1) candidacy for EBV placement and (2) target-lobe selection, were developed based on the normalised distributions of the four quantitative HRCT variables (fissure completeness score, per cent of voxel density with HU < −910 and HU < −950, and lobar volumes) across all five lung lobes. An external cohort of 50 patients (25 candidates and 25 non-candidates) was used to validate the prediction model. Results Performance measures of the training cohort demonstrated an area under the curve (AUC) of 0.91, accuracy of 81%, sensitivity of 93% and specificity of 78% compared with the multidisciplinary teams’ target lobe selection. The validation cohort demonstrated an AUC of 0.89, accuracy of 84%, sensitivity of 88% and specificity of 83% compared with the multidisciplinary team decision making. Conclusions Endobronchial valve lung volume reduction remains a potent palliative measure to improve quality of life in patients with hyperinflated emphysema. Our model for target lobe selection harnesses the multidisciplinary experience at a tertiary care centre to objectively select candidates and target lobes to assist clinicians’ decision making. Future studies investigating prediction of lobar collapse and functional improvement after target lobe selection using our model are needed.

Abstract Rationale Stereotactic body radiation therapy (SBRT) represents an effective therapeutic strategy for early-stage non–small cell lung cancer (NSCLC); however, local and systemic recurrences represent ongoing challenges. Computed tomography (CT) radiomics-based risk models can potentially be used to predict the risk of local recurrence on pretreatment CT scans. Objective Development of a radiomics model to predict local recurrence after SBRT in patients with NSCLC. Methods This single-institution study includes a retrospective case-control training set (20 patients with local recurrence and 40 control subjects) and an independent validation set (198 consecutive cases) of patients with early-stage NSCLC treated with SBRT. Tumors were semiautomatically segmented, and 102 quantitative radiomic features, including texture, landscape, spatial, nodule shape, and nodule surface features, were extracted. These features were included in three separate multivariable models to predict the risk of recurrence on the basis of pre-SBRT, post-SBRT, and the difference between the pre-SBRT and post-SBRT scans (Delta model). The pre-SBRT model was subsequently validated in an independent validation set. Results Thirteen independent variables were selected for the models using the Boruta algorithm. The sensitivity, specificity, and area under the curve of the pre-SBRT, post-SBRT, and Delta models were 85%, 90%, and 0.91; 85%, 92.5%, and 0.92; and 85%, 92.5%, and 0.94, respectively. The pre-SBRT model was validated in the independent validation set (area under the curve, 0.89; confidence interval, 0.83–0.92), because this model was believed to be the most useful to assist in individualized treatment planning. Conclusions Radiomic analysis facilitated the development of three high-performing models predicting local recurrence using either pre-SBRT CT, post-SBRT CT, or the change between these two. We successfully validated the most clinically relevant model, the pre-SBRT model. Although this model needs further validation, it may facilitate individualized surveillance, treatment planning, and selection of adjuvant therapy.

Abstract Introduction: While CT lung cancer screening reduces lung cancer-specific mortality, there are remaining challenges. Radiomic tools promiss to address these challenges, however, they are subject to interobserver variability if semi-automated segmentation techniques are used. Herein we report interobserver variability for two validated radiomic tools, BRODERS (Benign versus aggRessive nODule Evaluation using Radiomic Stratification) and CANARY (Computer-Aided Nodule Assessment and Risk Yield). Methods We retrospectively analyzed the CT images of 95 malignant lung nodules of the adenocarcinoma spectrum using BRODERS and CANARY. Cases were identified at Mayo Clinic (n = 45) and Vanderbilt University Medical Center and Nashville/Veteran Administration Tennessee Valley Health Care System (n = 50). Three observers with different training levels (medical student, internal medicine resident and thoracic radiology fellow) each performed lung nodule segmentation. All methods were carried out in accordance with relevant guidelines and regulations Results Interclass correlation coefficients (ICC) of 0.77, 0.98 and 0.97 for the average nodule volume, BRODERS cancer probability and Score Indicative of Lesion Aggression (SILA) which summarizes the distribution of the CANARY exemplars indicated good to excellent reliability, respectively. The dice similarity coefficient was 0.79 and 0.81 for the data sets from the two institutions. Conclusion: BRODERS and CANARY are robust radiomics tools with excellent interobserver variability. These tools are simple and reliable regardless the observer/operator’s level of training.

Purpose: Optimal strategies to detect early interstitial lung disease (ILD) are unknown. ILD is frequently subpleural in distribution and affects lung elasticity. Lung ultrasound surface wave elastography (LUSWE) is a noninvasive method of quantifying superficial lung tissue elastic properties. In LUWSE a handheld device applied at the intercostal space vibrates the chest at a set frequency, and the lung surface wave velocity is measured by an ultrasound probe 5 mm away in the same intercostal space. We explored LUWSE’s ability to detect ILD and correlated LUSWE velocity with physiological, quantitative, and visual radiologic features of subjects with known ILD and of healthy controls. Materials and Methods: Seventy-seven subjects with ILD, mostly caused by connective tissue disease, and 19 healthy controls were recruited. LUSWE was performed on all subjects in 3 intercostal lung regions bilaterally. Comparison of LUSWE velocities pulmonary function testing, visual assessment, and quantitative analysis of recent computed tomographic imaging with Computer-Aided Lung Informatics for Pathology Evaluation and Rating (CALIPER) software. Results: Sonographic velocities were higher in all lung regions for cases, with the greatest difference in the lateral lower lung. Median velocity in m/s was 5.84 versus 4.11 and 5.96 versus 4.27 (P<0.00001) for cases versus controls, left and right lateral lower lung zones, respectively. LUSWE velocity correlated negatively with vital capacity and positively with radiologist and CALIPER-detected interstitial abnormalities. Conclusions: LUSWE is a safe and noninvasive technique that shows high sensitivity to detect ILD and correlated with clinical, physiological, radiologic, and quantitative assessments of ILD. Prospective study in detecting ILD is indicated.

Abstract Rationale Quantitative computed tomographic (CT) measures of baseline disease severity might identify patients with idiopathic pulmonary fibrosis (IPF) with an increased mortality risk. We evaluated whether quantitative CT variables could act as a cohort enrichment tool in future IPF drug trials. Objectives To determine whether computer-derived CT measures, specifically measures of pulmonary vessel–related structures (VRSs), can better predict functional decline and survival in IPF and reduce requisite sample sizes in drug trial populations. Methods Patients with IPF undergoing volumetric noncontrast CT imaging at the Royal Brompton Hospital, London, and St. Antonius Hospital, Utrecht, were examined to identify pulmonary function measures (including FVC) and visual and computer-derived (CALIPER [Computer-Aided Lung Informatics for Pathology Evaluation and Rating] software) CT features predictive of mortality and FVC decline. The discovery cohort comprised 247 consecutive patients, with validation of results conducted in a separate cohort of 284 patients, all fulfilling drug trial entry criteria. Measurements and Main Results In the discovery and validation cohorts, CALIPER-derived features, particularly VRS scores, were among the strongest predictors of survival and FVC decline. CALIPER results were accentuated in patients with less extensive disease, outperforming pulmonary function measures. When used as a cohort enrichment tool, a CALIPER VRS score greater than 4.4% of the lung was able to reduce the requisite sample size of an IPF drug trial by 26%. Conclusions Our study has validated a new quantitative CT measure in patients with IPF fulfilling drug trial entry criteria—the VRS score—that outperformed current gold standard measures of outcome. When used for cohort enrichment in an IPF drug trial setting, VRS threshold scores can reduce a required IPF drug trial population size by 25%, thereby limiting prohibitive trial costs. Importantly, VRS scores identify patients in whom antifibrotic medication prolongs life and reduces FVC decline.

ABSTRACT Background and objective This study evaluated whether patients with combined pulmonary fibrosis and emphysema (CPFE) have an increased likelihood of pulmonary hypertension (PHT) when compared with idiopathic pulmonary fibrosis (IPF) patients without emphysema. Methods Two consecutive IPF populations having undergone transthoracic echocardiography were examined ( n = 223 and n = 162). Emphysema and interstitial lung disease (ILD) extent were quantified visually; ILD extent was also quantified by a software tool, CALIPER. Echocardiographic criteria categorized PHT risk. Results The prevalence of an increased PHT likelihood was 29% and 31% in each CPFE cohort. Survival at 12 months was 60% across both CPFE cohorts with no significantly worsened outcome identified when compared with IPF patients without emphysema. Using logistic regression models in both cohorts, total computed tomography (CT) disease extent (ILD and emphysema) predicted the likelihood of PHT. After adjustment for total disease extent, CPFE had no stronger association with PHT likelihood than IPF patients without emphysema. Conclusion Our findings indicate that the reported association between CPFE and PHT is explained by the summed baseline CT extents of ILD and emphysema. Once baseline severity is taken into account, CPFE is not selectively associated with a malignant microvascular phenotype, when compared with IPF patients without emphysema.

Abstract This study assessed the inter-user reproducibility of Computer-Aided Nodule Assessment and Risk Yield (CANARY), a novel analytical tool that risk stratifies lung adenocarcinomas (ADCs) according to defined computed tomography (CT) structural characteristics. CANARY detects nine distinct voxel signatures in ADCs based on standard CT imaging, thereby defining nodule characteristics correlating with patient outcomes, and corresponding to invasion or lepidic growth histologically. A software user segments the borders of each ADC prior to voxel analysis, introducing potential variability into the assessment. While CANARY is a promising method, it requires validation of the analytical variability between users and prediction of accuracy in an independent cohort. To evaluate the reproducibility of CANARY analysis, three independent users who are not part of the CANARY development team segmented and analyzed 50 biopsy-confirmed primary lung ADCs from Vanderbilt University Medical Center. The CT scans of ADCs were selected retrospectively based on the following criteria: ADC-histology proven on biopsy, diagnosed between 2009-2015, less than 3cm in greatest diameter, and stages IA-IV. Users followed a standard operating procedure established at the Mayo Clinic, and were blinded to clinical characteristics and patient outcomes. Results: To measure inter-user variability of ADC voxel composition, the intraclass correlation coefficient (ICC) was calculated based upon the percentage of each voxel subtype within the whole ADC. An ICC of 1 reflects high reproducibility between users. Amongst all 50 ADCs, the average ICC for all nine voxel types was 0.828 (95% CI 0.76, 0.895). The ICC of the four voxel types associated with invasive features on CT was 0.865 (95% CI 0.805, 0.924). ICCs were also calculated using a logarithmic transformation for data normalization, generating an ICC of 0.745 (95% CI 0.663, 0.826) for all nine voxel types, and an ICC of 0.995 (95% CI 0.993, 0.997) for the four voxel types associated with invasion. Conclusions: (1) CANARY analysis of lung ADC voxel signatures appears to be reproducible amongst users, making it a reliable tool for the evaluation of ADC voxel density. (2) Correlation of invasive signatures associated with more aggressive ADCs was nearly perfect amongst users. Additional validation metrics for CANARY with larger datasets are being evaluated, including the accuracy of tumor prognostic predictions between users and analysis of ADC datasets from other institutions. Citation Format: Erica C. Nakajima, Michael P. Frankland, Tucker Johnson, Sanja L. Antic, Ronald A. Karwoski, Bennett Landman, Heidi Chen, Ronald C. Walker, Brian J. Bartholmai, Tobias Peikert, Srinivasan Rajagopalan, Pierre P. Massion, Fabien Maldonado. Assessing the reproducibility of computer-aided nodule assessment and risk yield (CANARY) method to characterize lung adenocarcinomas [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 3723. doi:10.1158/1538-7445.AM2017-3723