Enhancing cancer prediction in challenging screen-detected incident lung nodules using time-series deep learning.

Lung cancer screening (LCS) using annual computed tomography (CT) scanning significantly reduces mortality by detecting cancerous lung nodules at an earlier stage. Deep learning algorithms can improve nodule malignancy risk stratification. However, they have typically been used to analyse single time point CT data when detecting malignant nodules on either baseline or incident CT LCS rounds. Deep learning algorithms have the greatest value in two aspects. These approaches have great potential in assessing nodule change across time-series CT scans where subtle changes may be challenging to identify using the human eye alone. Moreover, they could be targeted to detect nodules developing on incident screening rounds, where cancers are generally smaller and more challenging to detect confidently. Here, we show the performance of our Deep learning-based Computer-Aided Diagnosis model integrating Nodule and Lung imaging data with clinical Metadata Longitudinally (DeepCAD-NLM-L) for malignancy prediction. DeepCAD-NLM-L showed improved performance (AUC = 88%) against models utilizing single time-point data alone. DeepCAD-NLM-L also demonstrated comparable and complementary performance to radiologists when interpreting the most challenging nodules typically found in LCS programs. It also demonstrated similar performance to radiologists when assessed on out-of-distribution imaging dataset. The results emphasize the advantages of using time-series and multimodal analyses when interpreting malignancy risk in LCS.

Predicting programmed death-ligand 1 (PD-L1) expression with fluorine-18 fluorodeoxyglucose ([18F]FDG) positron emission tomography/computed tomography (PET/CT) metabolic parameters in resectable non-small cell lung cancer.

BACKGROUND: Programmed death-ligand 1 (PD-L1) expression is a predictive biomarker for immunotherapy in non-small cell lung cancer (NSCLC). PD-L1 and glucose transporter 1 expression are closely associated, and studies demonstrate correlation of PD-L1 with glucose metabolism. AIM: The aim of this study was to investigate the association of fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography ([18F]FDG-PET/CT) metabolic parameters with PD-L1 expression in primary lung tumour and lymph node metastases in resected NSCLC. METHODS: We conducted a retrospective analysis of 210 patients with node-positive resectable stage IIB-IIIB NSCLC. PD-L1 tumour proportion score (TPS) was determined using the DAKO 22C3 immunohistochemical assay. Semi-automated techniques were used to analyse pre-operative [18F]FDG-PET/CT images to determine primary and nodal metabolic parameter scores (including max, mean, peak and peak adjusted for lean body mass standardised uptake values (SUV), metabolic tumour volume (MTV), total lesional glycolysis (TLG) and SUV heterogeneity index (HISUV)). RESULTS: Patients were predominantly male (57%), median age 70 years with non-squamous NSCLC (68%). A majority had negative primary tumour PD-L1 (TPS < 1%; 53%). Mean SUVmax, SUVmean, SUVpeak and SULpeak values were significantly higher (p < 0.05) in those with TPS ≥ 1% in primary tumour (n = 210) or lymph nodes (n = 91). However, ROC analysis demonstrated only moderate separability at the 1% PD-L1 TPS threshold (AUCs 0.58-0.73). There was no association of MTV, TLG and HISUV with PD-L1 TPS. CONCLUSION: This study demonstrated the association of SUV-based [18F]FDG-PET/CT metabolic parameters with PD-L1 expression in primary tumour or lymph node metastasis in resectable NSCLC, but with poor sensitivity and specificity for predicting PD-L1 positivity ≥ 1%. CLINICAL RELEVANCE STATEMENT: Whilst SUV-based fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography metabolic parameters may not predict programmed death-ligand 1 positivity ≥ 1% in the primary tumour and lymph nodes of resectable non-small cell lung cancer independently, there is a clear association which warrants further investigation in prospective studies. TRIAL REGISTRATION: Non-applicable KEY POINTS: • Programmed death-ligand 1 immunohistochemistry has a predictive role in non-small cell lung cancer immunotherapy; however, it is both heterogenous and dynamic. • SUV-based fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography ([18F]FDG-PET/CT) metabolic parameters were significantly higher in primary tumour or lymph node metastases with positive programmed death-ligand 1 expression. • These SUV-based parameters could potentially play an additive role along with other multi-modal biomarkers in selecting patients within a predictive nomogram.

Patient Reidentification from Chest Radiographs: An Interpretable Deep Metric Learning Approach and Its Applications.

Purpose: To train an explainable deep learning model for patient reidentification in chest radiograph datasets and assess changes in model-perceived patient identity as a marker for emerging radiologic abnormalities in longitudinal image sets. Materials and Methods: This retrospective study used a set of 1 094 537 frontal chest radiographs and free-text reports from 259 152 patients obtained from six hospitals between 2006 and 2019, with validation on the public ChestX-ray14, CheXpert, and MIMIC-CXR datasets. A deep learning model was trained for patient reidentification and assessed on patient identity confirmation, retrieval of patient images from a database based on a query image, and radiologic abnormality prediction in longitudinal image sets. The representation learned was incorporated into a generative adversarial network, allowing visual explanations of the relevant features. Performance was evaluated with sensitivity, specificity, F1 score, Precision at 1, R-Precision, and area under the receiver operating characteristic curve (AUC) for normal and abnormal prediction. Results: Patient reidentification was achieved with a mean F1 score of 0.996 ± 0.001 (2 SD) on the internal test set (26 152 patients) and F1 scores of 0.947-0.993 on the external test data. Database retrieval yielded a mean Precision at 1 score of 0.976 ± 0.005 at 299 × 299 resolution on the internal test set and Precision at 1 scores between 0.868 and 0.950 on the external datasets. Patient sex, age, weight, and other factors were identified as key model features. The model achieved an AUC of 0.73 ± 0.01 for abnormality prediction versus an AUC of 0.58 ± 0.01 for age prediction error. Conclusion: The image features used by a deep learning patient reidentification model for chest radiographs corresponded to intuitive human-interpretable characteristics, and changes in these identifying features over time may act as markers for an emerging abnormality.Keywords: Conventional Radiography, Thorax, Feature Detection, Supervised Learning, Convolutional Neural Network, Principal Component Analysis Supplemental material is available for this article. © RSNA, 2023See also the commentary by Raghu and Lu in this issue.

Weakly supervised segmentation models as explainable radiological classifiers for lung tumour detection on CT images.

PURPOSE: Interpretability is essential for reliable convolutional neural network (CNN) image classifiers in radiological applications. We describe a weakly supervised segmentation model that learns to delineate the target object, trained with only image-level labels ("image contains object" or "image does not contain object"), presenting a different approach towards explainable object detectors for radiological imaging tasks. METHODS: A weakly supervised Unet architecture (WSUnet) was trained to learn lung tumour segmentation from image-level labelled data. WSUnet generates voxel probability maps with a Unet and then constructs an image-level prediction by global max-pooling, thereby facilitating image-level training. WSUnet's voxel-level predictions were compared to traditional model interpretation techniques (class activation mapping, integrated gradients and occlusion sensitivity) in CT data from three institutions (training/validation: n = 412; testing: n = 142). Methods were compared using voxel-level discrimination metrics and clinical value was assessed with a clinician preference survey on data from external institutions. RESULTS: Despite the absence of voxel-level labels in training, WSUnet's voxel-level predictions localised tumours precisely in both validation (precision: 0.77, 95% CI: [0.76-0.80]; dice: 0.43, 95% CI: [0.39-0.46]), and external testing (precision: 0.78, 95% CI: [0.76-0.81]; dice: 0.33, 95% CI: [0.32-0.35]). WSUnet's voxel-level discrimination outperformed the best comparator in validation (area under precision recall curve (AUPR): 0.55, 95% CI: [0.49-0.56] vs. 0.23, 95% CI: [0.21-0.25]) and testing (AUPR: 0.40, 95% CI: [0.38-0.41] vs. 0.36, 95% CI: [0.34-0.37]). Clinicians preferred WSUnet predictions in most instances (clinician preference rate: 0.72 95% CI: [0.68-0.77]). CONCLUSION: Weakly supervised segmentation is a viable approach by which explainable object detection models may be developed for medical imaging. CRITICAL RELEVANCE STATEMENT: WSUnet learns to segment images at voxel level, training only with image-level labels. A Unet backbone first generates a voxel-level probability map and then extracts the maximum voxel prediction as the image-level prediction. Thus, training uses only image-level annotations, reducing human workload. WSUnet's voxel-level predictions provide a causally verifiable explanation for its image-level prediction, improving interpretability. KEY POINTS: • Explainability and interpretability are essential for reliable medical image classifiers. • This study applies weakly supervised segmentation to generate explainable image classifiers. • The weakly supervised Unet inherently explains its image-level predictions at voxel level.

Prevalence and clinical characteristics of non-malignant CT detected incidental findings in the SUMMIT lung cancer screening cohort.

BACKGROUND: Pulmonary and extrapulmonary incidental findings are frequently identified on CT scans performed for lung cancer screening. Uncertainty regarding their clinical significance and how and when such findings should be reported back to clinicians and participants persists. We examined the prevalence of non-malignant incidental findings within a lung cancer screening cohort and investigated the morbidity and relevant risk factors associated with incidental findings. We quantified the primary and secondary care referrals generated by our protocol. METHODS: The SUMMIT study (NCT03934866) is a prospective observational cohort study to examine the performance of delivering a low-dose CT (LDCT) screening service to a high-risk population. Spirometry, blood pressure, height/weight and respiratory history were assessed as part of a Lung Health Check. Individuals at high risk of lung cancer were offered an LDCT and returned for two further annual visits. This analysis is a prospective evaluation of the standardised reporting and management protocol for incidental findings developed for the study on the baseline LDCT. RESULTS: In 11 115 participants included in this analysis, the most common incidental findings were coronary artery calcification (64.2%) and emphysema (33.4%). From our protocolised management approach, the number of participants requiring review for clinically relevant findings in primary care was 1 in 20, and the number potentially requiring review in secondary care was 1 in 25. CONCLUSIONS: Incidental findings are common in lung cancer screening and can be associated with reported symptoms and comorbidities. A standardised reporting protocol allows systematic assessment and standardises onward management.

The SUMMIT Study: Utilising a written 'Next Steps' information booklet to prepare participants for potential lung cancer screening results and follow-up.

OBJECTIVES: Low-Dose Computed Tomography (LDCT) screening for lung cancer can result in several potential outcomes of varying significance. Communication methods used in Lung Cancer Screening (LCS) programmes must, therefore, ensure that participants are prepared for the range of possible results and follow-up. Here, we assess perceptions of a written preparatory information booklet provided to participants in a large LCS cohort designed to convey this information. MATERIALS AND METHODS: All participants in the SUMMIT Study (NCT03934866) were provided with a results preparation information booklet, entitled 'The SUMMIT Study: Next Steps' at their baseline appointment which outlined potential results, their significance, and timelines for follow up. Results from the LDCT scan and Lung Health Check were subsequently sent by letter. Perceptions of this booklet were assessed among participants with indeterminate pulmonary findings when they attended a face-to-face appointment immediately before their three-month interval scan. Specifically, questions assessed the perceived usefulness of the booklet and the amount of information contained in it. RESULTS: 70.1% (n = 1,412/2,014) participants remembered receiving the booklet at their appointment. Of these participants, 72.0% (n = 1,017/1,412) found it quite or very useful and 68.0% (n = 960/1,412) reported that it contained the right amount of information. Older participants, those from the least deprived socioeconomic quintile and those of Black ethnicity were less likely to report finding the booklet either quite or very useful, or that it contained the right amount of information. Participants who remembered receiving the booklet were more likely to be satisfied with the process of results communication by letter. CONCLUSION: Providing written information that prepares participants for possible LDCT results and their significance appears to be a useful resource and a helpful adjunct to a written method of results communication for large scale LCS programmes.

Uptake of invitations to a lung health check offering low-dose CT lung cancer screening among an ethnically and socioeconomically diverse population at risk of lung cancer in the UK (SUMMIT): a prospective, longitudinal cohort study.

BACKGROUND: Lung cancer screening with low-dose CT reduces lung cancer mortality, but screening requires equitable uptake from candidates at high risk of lung cancer across ethnic and socioeconomic groups that are under-represented in clinical studies. We aimed to assess the uptake of invitations to a lung health check offering low-dose CT lung cancer screening in an ethnically and socioeconomically diverse cohort at high risk of lung cancer. METHODS: In this multicentre, prospective, longitudinal cohort study (SUMMIT), individuals aged 55-77 years with a history of smoking in the past 20 years were identified via National Health Service England primary care records at practices in northeast and north-central London, UK, using electronic searches. Eligible individuals were invited by letter to a lung health check offering lung cancer screening at one of four hospital sites, with non-responders re-invited after 4 months. Individuals were excluded if they had dementia or metastatic cancer, were receiving palliative care or were housebound, or declined research participation. The proportion of individuals invited who responded to the lung health check invitation by telephone was used to measure uptake. We used univariable and multivariable logistic regression analyses to estimate associations between uptake of a lung health check invitation and re-invitation of non-responders, adjusted for sex, age, ethnicity, smoking, and deprivation score. This study was registered prospectively with ClinicalTrials.gov, NCT03934866. FINDINGS: Between March 20 and Dec 12, 2019, the records of 2 333 488 individuals from 251 primary care practices across northeast and north-central London were screened for eligibility; 1 974 919 (84·6%) individuals were outside the eligible age range, 7578 (2·1%) had pre-existing medical conditions, and 11 962 (3·3%) had opted out of particpation in research and thus were not invited. 95 297 individuals were eligible for invitation, of whom 29 545 (31·0%) responded. Due to the COVID-19 pandemic, re-invitation letters were sent to only a subsample of 4594 non-responders, of whom 642 (14·0%) responded. Overall, uptake was lower among men than among women (odds ratio [OR] 0·91 [95% CI 0·88-0·94]; p<0·0001), and higher among older age groups (1·48 [1·42-1·54] among those aged 65-69 years vs those aged 55-59 years; p<0·0001), groups with less deprivation (1·89 [1·76-2·04] for the most vs the least deprived areas; p<0·0001), individuals of Asian ethnicity (1·14 [1·09-1·20] vs White ethnicity; p<0·0001), and individuals who were former smokers (1·89 [1·83-1·95] vs current smokers; p<0·0001). When ethnicity was subdivided into 16 groups, uptake was lower among individuals of other White ethnicity than among those with White British ethnicity (0·86 [0·83-0·90]), whereas uptake was higher among Chinese, Indian, and other Asian ethnicities than among those with White British ethnicity (1·33 [1·13-1·56] for Chinese ethnicity; 1·29 [1·19-1·40] for Indian ethnicity; and 1·19 [1·08-1·31] for other Asian ethnicity). INTERPRETATION: Inviting eligible adults for lung health checks in areas of socioeconomic and ethnic diversity should achieve favourable participation in lung cancer screening overall, but inequalities by smoking, deprivation, and ethnicity persist. Reminder and re-invitation strategies should be used to increase uptake and the equity of response. FUNDING: GRAIL.

Reporting and management of incidental lung findings on computed tomography: beyond lung nodules.

Non-nodular incidental lung findings can broadly be categorised as airway- or airspace-related abnormalities and diffuse parenchymal abnormalities. Airway-related abnormalities include bronchial dilatation and thickening, foci of low attenuation, emphysema, and congenital variants. Diffuse parenchymal abnormalities relate to the spectrum of diffuse parenchymal lung diseases cover a spectrum from interstitial lung abnormalities (ILAs) and pulmonary cysts to established diffuse parenchymal lung abnormalities such as the idiopathic interstitial pneumonias and cystic lung diseases. In this review, we discuss the main manifestations of these incidental findings, paying attention to their prevalence and importance, descriptors to use when reporting, the limits of what can be considered "normal", and conclude each section with some pragmatic reporting recommendations. We also highlight technical and patient factors which can lead to spurious abnormalities.