Automated Delineation of Lung Tumors from CT Images Using a Single Click Ensemble Segmentation Approach

Applying deep learning to segmentation of murine lung tumors in pre-clinical micro-computed tomography

Lung cancer remains a leading cause of cancer-related death, but scientists have made great strides in developing new treatments recently, partly owing to the use of genetically engineered mouse models (GEMMs). GEMM tumors represent a translational model that recapitulates human disease better than implanted models because tumors develop spontaneously in the lungs. However, detection of these tumors relies on in vivo imaging tools, specifically micro-Computed Tomography (micro-CT or µCT), and image analysis can be laborious with high inter-user variability. Here we present a deep learning model trained to perform fully automated segmentation of lung tumors without the interference of other soft tissues. Trained and tested on 100 3D µCT images (18,662 slices) that were manually segmented, the model demonstrated a high correlation to manual segmentations on the testing data (r2=0.99, DSC=0.78) and on an independent dataset (n = 12 3D scans or 2328 2D slices, r2=0.97, DSC=0.73). In a comparison against manual segmentation performed by multiple analysts, the model (r2=0.98, DSC=0.78) performed within inter-reader variability (r2=0.79, DSC=0.69) and close to intra-reader variability (r2=0.99, DSC=0.82), all while completing 5+ hours of manual segmentations in 1 minute. Finally, when applied to a real-world longitudinal study (n = 55 mice), the model successfully detected tumor progression over time and the differences in tumor burden between groups induced with different virus titers, aligning well with more traditional analysis methods. In conclusion, we have developed a deep learning model which can perform fast, accurate, and fully automated segmentation of µCT scans of murine lung tumors.

Evaluating clinical and radiomic features for predicting lung cancer recurrence pre- and post-tumor resection

Among patients with early-stage non-small cell lung cancer (NSCLC) undergoing surgical resection, identifying who is at high-risk of recurrence can inform clinical guidelines with respect to more aggressive follow-up and/or adjuvant therapy. While predicting recurrence based on pre-surgical resection data is ideal, clinically important pathological features are only evaluated postoperatively. Therefore, we developed two supervised classification models to assess the importance of pre- and post-surgical features for predicting 5-year recurrence. An integrated dataset was generated by combining clinical covariates and radiomic features calculated from pre-surgical computed tomography images. After removing correlated radiomic features, the SHapley Additive exPlanations (SHAP) method was used to measure feature importance and select relevant features. Binary classification was performed using a Support Vector Machine, followed by a feature ablation study assessing the impact of radiomic and clinical features. We demonstrate that the post-surgical model significantly outperforms the pre-surgical model in predicting lung cancer recurrence, with tumor pathological features and peritumoral radiomic features contributing significantly to the model's performance.

Artificial Intelligence and Machine Learning in Lung Cancer Screening

Recent advances in artificial intelligence and machine learning (AI/ML) hold substantial promise to address some of the current challenges in lung cancer screening and improve health equity. This article reviews the status and future directions of AI/ML tools in the lung cancer screening workflow, focusing on determining screening eligibility, radiation dose reduction and image denoising for low-dose chest computed tomography (CT), lung nodule detection, lung nodule classification, and determining optimal screening intervals. AI/ML tools can assess for chronic diseases on CT, which creates opportunities to improve population health through opportunistic screening.

An Automatic Segmentation Method for Lung Tumor Based on Improved Region Growing Algorithm

In medical image processing, accurate segmentation of lung tumors is very important. Computer-aided accurate segmentation can effectively assist doctors in surgery planning and treatment decisions. Although the accurate segmentation results of lung tumors can provide a reliable basis for clinical treatment, the key to obtaining accurate segmentation results is how to improve the segmentation performance of the algorithm. We propose an automatic segmentation method for lung tumors based on an improved region growing algorithm, which uses the prior information on lung tumors to achieve an automatic selection of the initial seed point. The proposed method includes a seed point expansion mechanism and an automatic threshold update mechanism and takes the combination of multiple segmentation results as the final segmentation result. In the lung image database consortium (LIDC-IDRI) dataset, we designed 10 experiments to test the proposed method and compare it with 4 popular segmentation methods. The experimental results show that the average dice coefficient obtained by the proposed method is 0.936 ± 0.027, and the average Jaccard distance is 0.114 ± 0.049. The average dice coefficient obtained by the proposed method is 0.107, 0.053, 0.040, and 0.156, higher than that of the other four methods, respectively. This study proves that the proposed method can automatically segment lung tumors in CT slices and has suitable segmentation performance.

Application of Machine Learning and Deep Learning Techniques for COVID-19 Screening Using Radiological Imaging: A Comprehensive Review

Lung, being one of the most important organs in human body, is often affected by various SARS diseases, among which COVID-19 has been found to be the most fatal disease in recent times. In fact, SARS-COVID 19 led to pandemic that spreads fast among the community causing respiratory problems. Under such situation, radiological imaging-based screening [mostly chest X-ray and computer tomography (CT) modalities] has been performed for rapid screening of the disease as it is a non-invasive approach. Due to scarcity of physician/chest specialist/expert doctors, technology-enabled disease screening techniques have been developed by several researchers with the help of artificial intelligence and machine learning (AI/ML). It can be remarkably observed that the researchers have introduced several AI/ML/DL (deep learning) algorithms for computer-assisted detection of COVID-19 using chest X-ray and CT images. In this paper, a comprehensive review has been conducted to summarize the works related to applications of AI/ML/DL for diagnostic prediction of COVID-19, mainly using X-ray and CT images. Following the PRISMA guidelines, total 265 articles have been selected out of 1715 published articles till the third quarter of 2021. Furthermore, this review summarizes and compares varieties of ML/DL techniques, various datasets, and their results using X-ray and CT imaging. A detailed discussion has been made on the novelty of the published works, along with advantages and limitations.

A transfer learning based deep learning model to diagnose covid-19 CT scan images

To save the life of human beings during the pandemic conditions we need an effective automated method to deal with this situation. In pandemic conditions when the available resources becomes insufficient to handle the patient's load, then we needed some fast and reliable method which analyse the patient medical data with high efficiency and accuracy within time limitations. In this manuscript, an effective and efficient method is proposed for exact diagnosis of the patient whether it is coronavirus disease-2019 (covid-19) positive or negative with the help of deep learning. To find the correct diagnosis with high accuracy we use pre-processed segmented images for the analysis with deep learning. In the first step the X-ray image or computed tomography (CT) of a covid-19 infected person is analysed with various schemes of image segmentation like simple thresholding at 0.3, simple thresholding at 0.6, multiple thresholding (between 26-230) and Otsu's algorithm. On comparative analysis of all these methods, it is found that the Otsu's algorithm is a simple and optimum scheme to improve the segmented outcome of binary image for the diagnosis point of view. Otsu's segmentation scheme gives more precise values in comparison to other methods on the scale of various image quality parameters like accuracy, sensitivity, f-measure, precision, and specificity. For image classification here we use Resnet-50, MobileNet and VGG-16 models of deep learning which gives accuracy 70.24%, 72.95% and 83.18% respectively with non-segmented CT scan images and 75.08%, 80.12% and 99.28% respectively with Otsu's segmented CT scan images. On a comparative study we find that the VGG-16 models with CT scan image segmented with Otsu's segmentation gives very high accuracy of 99.28%. On the basis of the diagnosis of the patient firstly we go for an arterial blood gas (ABG) analysis and then on the behalf of this diagnosis and ABG report, the severity level of the patient can be decided and according to this severity level, proper treatment protocols can be followed immediately to save the patient's life. Compared with the existing works, our deep learning based novel method reduces the complexity, takes much less time and has a greater accuracy for exact diagnosis of coronavirus disease-2019 (covid-19).

Positron Emission Tomography Image Segmentation Based on Atanassov’s Intuitionistic Fuzzy Sets

In this paper, we present an approach to fully automate tumor delineation in positron emission tomography (PET) images. PET images play a major role in medicine for in vivo imaging in oncology (PET images are used to evaluate oncology patients, detecting emitted photons from a radiotracer localized in abnormal cells). PET image tumor delineation plays a vital role both in pre- and post-treatment stages. The low spatial resolution and high noise characteristics of PET images increase the challenge in PET image segmentation. Despite the difficulties and known limitations, several image segmentation approaches have been proposed. This paper introduces a new unsupervised approach to perform tumor delineation in PET images using Atanassov’s intuitionistic fuzzy sets (A-IFSs) and restricted dissimilarity functions. Moreover, the implementation of this methodology is presented and tested against other existing methodologies. The proposed algorithm increases the accuracy of tumor delineation in PET images, and the experimental results show that the proposed method outperformed all methods tested.

EGFR Mutation Status and Subtypes Predicted by CT-Based 3D Radiomic Features in Lung Adenocarcinoma

Objective

In this study, we aim to establish a non-invasive tool to predict epidermal growth factor receptor (EGFR) mutation status and subtypes based on radiomic features of computed tomography (CT).

Methods

A total of 233 lung adenocarcinoma patients were investigated and randomly divided into the training and test cohorts. In this study, 2300 radiomic features were extracted from original and filtered (Exponential, Laplacian of Gaussian, Logarithm, Gabor, Wavelet) CT images. The radiomic features were divided into four categories, including histogram, volumetric, morphologic, and texture features. An RF-BFE algorithm was developed to select the features for building the prediction models. Clinicopathological features (including age, gender, smoking status, TNM staging, maximum diameter, location, and growth pattern) were combined to establish an integrated model with radiomic features. ROC curve and AUC quantified the effectiveness of the predictor of EGFR mutation status and subtypes.

Results

A set of 10 features were selected to predict EGFR mutation status between EGFR mutant and wild type, while 9 selected features were used to predict mutation subtypes between exon 19 deletion and exon 21 L858R mutation. To predict the EGFR mutation status, the AUC of the training cohort was 0.778 and the AUC of the test cohort was 0.765. To predict the EGFR mutation subtypes, the AUC of training cohort was 0.725 and the AUC of test cohort was 0.657. The integrated model showed the most optimal predictive performance with EGFR mutation status (AUC = 0.870 and 0.759) and subtypes (AUC = 0.797 and 0.554) in the training and test cohorts.

Conclusion

CT-based radiomic features can extract information on tumor heterogeneity in lung adenocarcinoma. In addition, we have established a radiomic model and an integrated model to non-invasively predict the EGFR mutation status and subtypes of lung adenocarcinoma, which is conducive to saving clinical costs and guiding targeted therapy.

Teacher-student approach for lung tumor segmentation from mixed-supervised datasets

PURPOSE

Cancer is among the leading causes of death in the developed world, and lung cancer is the most lethal type. Early detection is crucial for better prognosis, but can be resource intensive to achieve. Automating tasks such as lung tumor localization and segmentation in radiological images can free valuable time for radiologists and other clinical personnel. Convolutional neural networks may be suited for such tasks, but require substantial amounts of labeled data to train. Obtaining labeled data is a challenge, especially in the medical domain.

METHODS

This paper investigates the use of a teacher-student design to utilize datasets with different types of supervision to train an automatic model performing pulmonary tumor segmentation on computed tomography images. The framework consists of two models: the student that performs end-to-end automatic tumor segmentation and the teacher that supplies the student additional pseudo-annotated data during training.

RESULTS

Using only a small proportion of semantically labeled data and a large number of bounding box annotated data, we achieved competitive performance using a teacher-student design. Models trained on larger amounts of semantic annotations did not perform better than those trained on teacher-annotated data. Our model trained on a small number of semantically labeled data achieved a mean dice similarity coefficient of 71.0 on the MSD Lung dataset.

CONCLUSIONS

Our results demonstrate the potential of utilizing teacher-student designs to reduce the annotation load, as less supervised annotation schemes may be performed, without any real degradation in segmentation accuracy.

Radiomics in medical imaging: pitfalls and challenges in clinical management

BACKGROUND

Radiomics and radiogenomics are two words that recur often in language of radiologists, nuclear doctors and medical physicists especially in oncology field. Radiomics is the technique of medical images analysis to extract quantitative data that are not detected by human eye.

METHODS

This article is a narrative review on Radiomics in Medical Imaging. In particular, the review exposes the process, the limitations related to radiomics, and future prospects are discussed.

RESULTS

Several studies showed that radiomics is very promising. However, there were some critical issues: poor standardization and generalization of radiomics results, data-quality control, repeatability, reproducibility, database balancing and issues related to model overfitting.

CONCLUSIONS

Radiomics procedure should made considered all pitfalls and challenges to obtain robust and reproducible results that could be generalized in other patients cohort.

Deep Residual Separable Convolutional Neural Network for lung tumor segmentation

Lung cancer is one of the deadliest types of cancers. Computed Tomography (CT) is a widely used technique to detect tumors present inside the lungs. Delineation of such tumors is particularly essential for analysis and treatment purposes. With the advancement in hardware technologies, Machine Learning and Deep Learning methods are outperforming the traditional methods in the field of medical imaging. In order to delineate lung cancer tumors, we have proposed a deep learning-based methodology which includes a maximum intensity projection based pre-processing method, two novel deep learning networks and an ensemble strategy. The two proposed networks named Deep Residual Separable Convolutional Neural Network 1 and 2 (DRS-CNN1 and DRS-CNN2) achieved better performance over the state-of-the-art U-net network and other segmentation networks. For fair comparison, we have evaluated the performances of all networks on Medical Segmentation Decathlon (MSD) and StructSeg 2019 datasets. The DRS-CNN2 achieved a mean Dice Similarity Coefficient (DSC) of 0.649, mean 95 Hausdorff Distance (HD95) of 18.26, mean Sensitivity 0.737 and a mean Precision of 0.765 on independent test sets.

Water Cycle Bat Algorithm and Dictionary-Based Deformable Model for Lung Tumor Segmentation

Among the different types of cancers, lung cancer is one of the widespread diseases which causes the highest number of deaths every year. The early detection of lung cancer is very essential for increasing the survival rate in patients. Although computed tomography (CT) is the preferred choice for lungs imaging, sometimes CT images may produce less tumor visibility regions and unconstructive rates in tumor portions. Hence, the development of an efficient segmentation technique is necessary. In this paper, water cycle bat algorithm- (WCBA-) based deformable model approach is proposed for lung tumor segmentation. In the preprocessing stage, a median filter is used to remove the noise from the input image and to segment the lung lobe regions, and Bayesian fuzzy clustering is applied. In the proposed method, deformable model is modified by the dictionary-based algorithm to segment the lung tumor accurately. In the dictionary-based algorithm, the update equation is modified by the proposed WCBA and is designed by integrating water cycle algorithm (WCA) and bat algorithm (BA).

Lung tumor segmentation in 4D CT images using motion convolutional neural networks

PURPOSE

Manual delineation on all breathing phases of lung cancer 4D CT image datasets can be challenging, exhaustive, and prone to subjective errors because of both the large number of images in the datasets and variations in the spatial location of tumors secondary to respiratory motion. The purpose of this work is to present a new deep learning-based framework for fast and accurate segmentation of lung tumors on 4D CT image sets.

METHODS

The proposed DL framework leverages motion region convolutional neural network (R-CNN). Through integration of global and local motion estimation network architectures, the network can learn both major and minor changes caused by tumor motion. Our network design first extracts tumor motion information by feeding 4D CT images with consecutive phases into an integrated backbone network architecture, locating volume-of-interest (VOIs) via a regional proposal network and removing irrelevant information via a regional convolutional neural network. Extracted motion information is then advanced into the subsequent global and local motion head network architecture to predict corresponding deformation vector fields (DVFs) and further adjust tumor VOIs. Binary masks of tumors are then segmented within adjusted VOIs via a mask head. A self-attention strategy is incorporated in the mask head network to remove any noisy features that might impact segmentation performance. We performed two sets of experiments. In the first experiment, a five-fold cross-validation on 20 4D CT datasets, each consisting of 10 breathing phases (i.e., 200 3D image volumes in total). The network performance was also evaluated on an additional unseen 200 3D images volumes from 20 hold-out 4D CT datasets. In the second experiment, we trained another model with 40 patients' 4D CT datasets from experiment 1 and evaluated on additional unseen nine patients' 4D CT datasets. The Dice similarity coefficient (DSC), center of mass distance (CMD), 95th percentile Hausdorff distance (HD95 ), mean surface distance (MSD), and volume difference (VD) between the manual and segmented tumor contour were computed to evaluate tumor detection and segmentation accuracy. The performance of our method was quantitatively evaluated against four different methods (VoxelMorph, U-Net, network without global and local networks, and network without attention gate strategy) across all evaluation metrics through a paired t-test.

RESULTS

The proposed fully automated DL method yielded good overall agreement with the ground truth for contoured tumor volume and segmentation accuracy. Our model yielded significantly better values of evaluation metrics (p < 0.05) than all four competing methods in both experiments. On hold-out datasets of experiment 1 and 2, our method yielded DSC of 0.86 and 0.90 compared to 0.82 and 0.87, 0.75 and 0.83, 081 and 0.89, and 0.81 and 0.89 yielded by VoxelMorph, U-Net, network without global and local networks, and networks without attention gate strategy. Tumor VD between ground truth and our method was the smallest with the value of 0.50 compared to 0.99, 1.01, 0.92, and 0.93 for between ground truth and VoxelMorph, U-Net, network without global and local networks, and networks without attention gate strategy, respectively.

CONCLUSIONS

Our proposed DL framework of tumor segmentation on lung cancer 4D CT datasets demonstrates a significant promise for fully automated delineation. The promising results of this work provide impetus for its integration into the 4D CT treatment planning workflow to improve the accuracy and efficiency of lung radiotherapy.

Mouse lung automated segmentation tool for quantifying lung tumors after micro-computed tomography

Unlike the majority of cancers, survival for lung cancer has not shown much improvement since the early 1970s and survival rates remain low. Genetically engineered mice tumor models are of high translational relevance as we can generate tissue specific mutations which are observed in lung cancer patients. Since these tumors cannot be detected and quantified by traditional methods, we use micro-computed tomography imaging for longitudinal evaluation and to measure response to therapy. Conventionally, we analyze microCT images of lung cancer via a manual segmentation. Manual segmentation is time-consuming and sensitive to intra- and inter-analyst variation. To overcome the limitations of manual segmentation, we set out to develop a fully-automated alternative, the Mouse Lung Automated Segmentation Tool (MLAST). MLAST locates the thoracic region of interest, thresholds and categorizes the lung field into three tissue categories: soft tissue, intermediate, and lung. An increase in the tumor burden was measured by a decrease in lung volume with a simultaneous increase in soft and intermediate tissue quantities. MLAST segmentation was validated against three methods: manual scoring, manual segmentation, and histology. MLAST was applied in an efficacy trial using a Kras/Lkb1 non-small cell lung cancer model and demonstrated adequate precision and sensitivity in quantifying tumor growth inhibition after drug treatment. Implementation of MLAST has considerably accelerated the microCT data analysis, allowing for larger study sizes and mid-study readouts. This study illustrates how automated image analysis tools for large datasets can be used in preclinical imaging to deliver high throughput and quantitative results.

Automatic COVID-19 lung infected region segmentation and measurement using CT-scans images

History shows that the infectious disease (COVID-19) can stun the world quickly, causing massive losses to health, resulting in a profound impact on the lives of billions of people, from both a safety and an economic perspective, for controlling the COVID-19 pandemic. The best strategy is to provide early intervention to stop the spread of the disease. In general, Computer Tomography (CT) is used to detect tumors in pneumonia, lungs, tuberculosis, emphysema, or other pleura (the membrane covering the lungs) diseases. Disadvantages of CT imaging system are: inferior soft tissue contrast compared to MRI as it is X-ray-based Radiation exposure. Lung CT image segmentation is a necessary initial step for lung image analysis. The main challenges of segmentation algorithms exaggerated due to intensity in-homogeneity, presence of artifacts, and closeness in the gray level of different soft tissue. The goal of this paper is to design and evaluate an automatic tool for automatic COVID-19 Lung Infection segmentation and measurement using chest CT images. The extensive computer simulations show better efficiency and flexibility of this end-to-end learning approach on CT image segmentation with image enhancement comparing to the state of the art segmentation approaches, namely GraphCut, Medical Image Segmentation (MIS), and Watershed. Experiments performed on COVID-CT-Dataset containing (275) CT scans that are positive for COVID-19 and new data acquired from the EL-BAYANE center for Radiology and Medical Imaging. The means of statistical measures obtained using the accuracy, sensitivity, F-measure, precision, MCC, Dice, Jacquard, and specificity are 0.98, 0.73, 0.71, 0.73, 0.71, 0.71, 0.57, 0.99 respectively; which is better than methods mentioned above. The achieved results prove that the proposed approach is more robust, accurate, and straightforward.

The value of asphericity derived from T1-weighted MR in differentiating intraparenchymal ring-enhancing lesions-comparison of glioblastomas and brain abscesses

BACKGROUND

Both brain abscess(BA)and glioblastoma (GBM) are common causative pathologies of intraparenchymal ring-enhancing lesions. Advanced MR sequences such as diffusion weighted image (DWI) were often used to increase distinguishability of both entities.

PURPOSE

To evaluate the value of asphericity (ASP) from conventional T1-weighted MR images in differentiating BA from morphologically similar ring-enhancing GBM.

MATERIAL AND METHODS

Twenty-one BA and twenty-nine GBM were retrospectively included in this study. Each region of interest (ROI) was delineated twice with the software of ITK-SNAP on the contrast-enhanced T1 images by two observers. ASP was calculated to define the relative deviation of the ROI's shape from a sphere. Intraclass correlation coefficients (ICC) for inter-observer and intra-observer were calculated. The diagnostic capabilities of ASP and conventional volume (VOL) of ROI were evaluated with receiver operating characteristic (ROC) curve analysis. In addition, areas under the ROC curves of ASP and VOL were compared.

RESULTS

ICC of intra-observer and inter-observer were 0.99 (95% confidence interval, [CI] 0.97-0.99) and 0.98 (0.95-0.99), respectively. Both ASP and VOL showed significant difference between BA and GBM. The mean ASP values for BA and GBM were 66.3±7.8 and 14.7±1.8, respectively. The mean VOL value of BA was also larger than that of GBM (47.2±7.4 vs. 20.7±1.5 mm³). The mean AUC of ASP and VOL were 0.977 (95% CI 0.944-1) and 0.86 (95% CI 0.746-0.974), respectively. The AUC of ASP was significantly higher than that of VOL (p=0.04). The optimal cut point values of ASP and VOL were 24.39 and 24.86 mm³, respectively.

CONCLUSIONS

ASP derived from routine MRI is useful in differentiating BA from GBM.

A Semi-Automatic Method To Segment The Left Atrium in MR Volumes With Varying Slice Numbers

Atrial fibrillation (AF) is the most common sustained arrhythmia and is associated with dramatic increases in mortality and morbidity. Atrial cine MR images are increasingly used in the management of this condition, but there are few specific tools to aid in the segmentation of such data. Some characteristics of atrial cine MR (thick slices, variable number of slices in a volume) preclude the direct use of traditional segmentation tools. When combined with scarcity of labelled data and similarity of the intensity and texture of the left atrium (LA) to other cardiac structures, the segmentation of the LA in CINE MRI becomes a difficult task. To deal with these challenges, we propose a semi-automatic method to segment the left atrium (LA) in MR images, which requires an initial user click per volume. The manually given location information is used to generate a chamber location map to roughly locate the LA, which is then used as an input to a deep network with slightly over 0.5 million parameters. A tracking method is introduced to pass the location information across a volume and to remove unwanted structures in segmentation maps. According to the results of our experiments conducted in an in-house MRI dataset, the proposed method outperforms the U-Net [1] with a margin of 20 mm on Hausdorff distance and 0.17 on Dice score, with limited manual interaction.

Reinventing radiation therapy with machine learning and imaging bio-markers (radiomics): State-of-the-art, challenges and perspectives

Radiation therapy is a pivotal cancer treatment that has significantly progressed over the last decade due to numerous technological breakthroughs. Imaging is now playing a critical role on deployment of the clinical workflow, both for treatment planning and treatment delivery. Machine-learning analysis of predefined features extracted from medical images, i.e. radiomics, has emerged as a promising clinical tool for a wide range of clinical problems addressing drug development, clinical diagnosis, treatment selection and implementation as well as prognosis. Radiomics denotes a paradigm shift redefining medical images as a quantitative asset for data-driven precision medicine. The adoption of machine-learning in a clinical setting and in particular of radiomics features requires the selection of robust, representative and clinically interpretable biomarkers that are properly evaluated on a representative clinical data set. To be clinically relevant, radiomics must not only improve patients' management with great accuracy but also be reproducible and generalizable. Hence, this review explores the existing literature and exposes its potential technical caveats, such as the lack of quality control, standardization, sufficient sample size, type of data collection, and external validation. Based upon the analysis of 165 original research studies based on PET, CT-scan, and MRI, this review provides an overview of new concepts, and hypotheses generating findings that should be validated. In particular, it describes evolving research trends to enhance several clinical tasks such as prognostication, treatment planning, response assessment, prediction of recurrence/relapse, and prediction of toxicity. Perspectives regarding the implementation of an AI-based radiotherapy workflow are presented.

CT Radiomics Signature of Tumor and Peritumoral Lung Parenchyma to Predict Nonsmall Cell Lung Cancer Postsurgical Recurrence Risk

RATIONALE AND OBJECTIVES

To estimate recurrence risk after surgery in nonsmall cell lung cancer (NSCLC) patients by employing tumoral and peritumoral radiomics analysis.

MATERIALS AND METHODS

One-hundred twenty-four surgically treated stage IA-IIB NSCLC patients' data from 2008 to 2013 were retrospectively collected. Patient outcome was defined as local recurrence (LR), distant metastasis (DM), and (sum of LR and DM) total recurrence (TR) at follow-up. Volumetric region of interests (ROIs) were drawn for the tumor, peritumoral lung parenchyma (2 cm around the tumor) and involved lobe on CT images. Ninety-four (morphological, first-order, textural, fractal-based) radiomics features were extracted from the ROIs and datasets were created from single or combined ROIs. Predictive models were built with radiomics signature (RS) and clinicopathological data, and the area under the curve (AUC) was used to evaluate the performance. Radiomics score was calculated with the best models' feature coefficients, low- and high-risk groups of patients defined accordingly. Kaplan-Meier curves were built, and the log-rank test was used for comparison among low- and high-risk groups. Differences in recurrence risk among the two risk groups were calculated (chi-square test).

RESULTS

Fifty-six patients developed TR (25 LR, 31 DM). The tumor-node-metastasis (TNM) stage recurrence predictability (AUC_TR 0.680; AUC_DM 0.672; AUC_LR 0.580) was substantially improved when RS was added to the predictive model (AUC_TR 0.760; AUC_DM 0.759; AUC_LR 0.750). Seventy-five percent of high-risk patients developed TR. Recurrence risk of the high-risk group was 16-fold higher than that of the low-risk group (p < 0.001).

CONCLUSION

Combination of the tumoral and peritumoral RS with TNM staging system outperformed TNM staging alone in individualized recurrence risk estimation of patients with surgically treated NSCLC.

Spline curve deformation model with prior shapes for identifying adhesion boundaries between large lung tumors and tissues around lungs in CT images

PURPOSE

Automated segmentation of lung tumors attached to anatomic structures such as the chest wall or mediastinum remains a technical challenge because of the similar Hounsfield units of these structures. To address this challenge, we propose herein a spline curve deformation model that combines prior shapes to correct large spatially contiguous errors (LSCEs) in input shapes derived from image-appearance cues.The model is then used to identify the adhesion boundaries between large lung tumors and tissue around the lungs.

METHODS

The deformation of the whole curve is driven by the transformation of the control points (CPs) of the spline curve, which are influenced by external and internal forces. The external force drives the model to fit the positions of the non-LSCEs of the input shapes while the internal force ensures the local similarity of the displacements of the neighboring CPs. The proposed model corrects the gross errors in the lung input shape caused by large lung tumors, where the initial lung shape for the model is inferred from the training shapes by shape group-based sparse prior information and the input lung shape is inferred by adaptive-thresholding-based segmentation followed by morphological refinement.

RESULTS

The accuracy of the proposed model is verified by applying it to images of lungs with either moderate large-sized (ML) tumors or giant large-sized (GL) tumors. The quantitative results in terms of the averages of the dice similarity coefficient (DSC) and the Jaccard similarity index (SI) are 0.982 ± 0.006 and 0.965 ± 0.012 for segmentation of lungs adhered by ML tumors, and 0.952 ± 0.048 and 0.926 ± 0.059 for segmentation of lungs adhered by GL tumors, which give 0.943 ± 0.021 and 0.897 ± 0.041 for segmentation of the ML tumors, and 0.907 ± 0.057 and 0.888 ± 0.091 for segmentation of the GL tumors, respectively. In addition, the bidirectional Hausdorff distances are 5.7 ± 1.4 and 11.3 ± 2.5 mm for segmentation of lungs with ML and GL tumors, respectively.

CONCLUSIONS

When combined with prior shapes, the proposed spline curve deformation can deal with large spatially consecutive errors in object shapes obtained from image-appearance information. We verified this method by applying it to the segmentation of lungs with large tumors adhered to the tissue around the lungs and the large tumors. Both the qualitative and quantitative results are more accurate and repeatable than results obtained with current state-of-the-art techniques.

Stability and reproducibility of computed tomography radiomic features extracted from peritumoral regions of lung cancer lesions

PURPOSE

Recent efforts have demonstrated that radiomic features extracted from the peritumoral region, the area surrounding the tumor parenchyma, have clinical utility in various cancer types. However, as like any radiomic features, peritumoral features could also be unstable and/or nonreproducible. Hence, the purpose of this study was to assess the stability and reproducibility of computed tomography (CT) radiomic features extracted from the peritumoral regions of lung lesions where stability was defined as the consistency of a feature by different segmentations, and reproducibility was defined as the consistency of a feature to different image acquisitions.

METHODS

Stability was measured utilizing the "moist run" dataset and reproducibility was measured utilizing the Reference Image Database to Evaluate Therapy Response test-retest dataset. Peritumoral radiomic features were extracted from incremental distances of 3-12 mm outside the tumor segmentation. A total of 264 statistical, histogram, and texture radiomic features were assessed from the selected peritumoral region-of-interests (ROIs). All features (except wavelet texture features) were extracted using standardized algorithms defined by the Image Biomarker Standardisation Initiative. Stability and reproducibility of features were assessed using the concordance correlation coefficient. The clinical utility of stable and reproducible peritumoral features was tested in three previously published lung cancer datasets using overall survival as the endpoint.

RESULTS

Features found to be stable and reproducible, regardless of the peritumoral distances, included statistical, histogram, and a subset of texture features suggesting that these features are less affected by changes (e.g., size or shape) of the peritumoral region due to different segmentations and image acquisitions. The stability and reproducibility of Laws and wavelet texture features were inconsistent across all peritumoral distances. The analyses also revealed that a subset of features were consistently stable irrespective of the initial parameters (e.g., seed point) for a given segmentation algorithm. No significant differences were found in stability for features that were extracted from ROIs bounded by a lung parenchyma mask versus ROIs that were not bounded by a lung parenchyma mask (i.e., peritumoral regions that extended outside of lung parenchyma). After testing the clinical utility of peritumoral features, stable and reproducible features were shown to be more likely to create repeatable models than unstable and nonreproducible features.

CONCLUSIONS

This study identified a subset of stable and reproducible CT radiomic features extracted from the peritumoral region of lung lesions. The stable and reproducible features identified in this study could be applied to a feature selection pipeline for CT radiomic analyses. According to our findings, top performing features in survival models were more likely to be stable and reproducible hence, it may be best practice to utilize them to achieve repeatable studies and reduce the chance of overfitting.

Automatic glioma segmentation based on adaptive superpixel

Background

The automatic glioma segmentation is of great significance for clinical practice. This study aims to propose an automatic method based on superpixel for glioma segmentation from the T2 weighted Magnetic Resonance Imaging.

Methods

The proposed method mainly includes three steps. First, we propose an adaptive superpixel generation algorithm based on simple linear iterative clustering version with 0 parameter (ASLIC0). This algorithm can acquire a superpixel image with fewer superpixels and better fit the boundary of region of interest (ROI) by automatically selecting the optimal number of superpixels. Second, we compose a training set by calculating the statistical, texture, curvature and fractal features for each superpixel. Third, Support Vector Machine (SVM) is used to train classification model based on the features of the second step.

Results

The experimental results on Multimodal Brain Tumor Image Segmentation Benchmark 2017 (BraTS2017) show that the proposed method has good segmentation performance. The average Dice, Hausdorff distance, sensitivity, and specificity for the segmented tumor against the ground truth are 0.8492, 3.4697 pixels, 81.47, and 99.64%, respectively. The proposed method shows good stability on high- and low-grade glioma samples. Comparative experimental results show that the proposed method has superior performance.

Conclusions

This provides a close match to expert delineation across all grades of glioma, leading to a fast and reproducible method of glioma segmentation.

Quantitative Imaging features Improve Discrimination of Malignancy in Pulmonary nodules

Pulmonary nodules are frequently detected radiological abnormalities in lung cancer screening. Nodules of the highest- and lowest-risk for cancer are often easily diagnosed by a trained radiologist there is still a high rate of indeterminate pulmonary nodules (IPN) of unknown risk. Here, we test the hypothesis that computer extracted quantitative features (“radiomics”) can provide improved risk-assessment in the diagnostic setting. Nodules were segmented in 3D and 219 quantitative features are extracted from these volumes. Using these features novel malignancy risk predictors are formed with various stratifications based on size, shape and texture feature categories. We used images and data from the National Lung Screening Trial (NLST), curated a subset of 479 participants (244 for training and 235 for testing) that included incident lung cancers and nodule-positive controls. After removing redundant and non-reproducible features, optimal linear classifiers with area under the receiver operator characteristics (AUROC) curves were used with an exhaustive search approach to find a discriminant set of image features, which were validated in an independent test dataset. We identified several strong predictive models, using size and shape features the highest AUROC was 0.80. Using non-size based features the highest AUROC was 0.85. Combining features from all the categories, the highest AUROC were 0.83.

Revealing Tumor Habitats from Texture Heterogeneity Analysis for Classification of Lung Cancer Malignancy and Aggressiveness

We propose an approach for characterizing structural heterogeneity of lung cancer nodules using Computed Tomography Texture Analysis (CTTA). Measures of heterogeneity were used to test the hypothesis that heterogeneity can be used as predictor of nodule malignancy and patient survival. To do this, we use the National Lung Screening Trial (NLST) dataset to determine if heterogeneity can represent differences between nodules in lung cancer and nodules in non-lung cancer patients. 253 participants are in the training set and 207 participants in the test set. To discriminate cancerous from non-cancerous nodules at the time of diagnosis, a combination of heterogeneity and radiomic features were evaluated to produce the best area under receiver operating characteristic curve (AUROC) of 0.85 and accuracy 81.64%. Second, we tested the hypothesis that heterogeneity can predict patient survival. We analyzed 40 patients diagnosed with lung adenocarcinoma (20 short-term and 20 long-term survival patients) using a leave-one-out cross validation approach for performance evaluation. A combination of heterogeneity features and radiomic features produce an AUROC of 0.9 and an accuracy of 85% to discriminate long- and short-term survivors.

Robust Pulmonary Nodule Segmentation in CT Image for Juxta-pleural and Juxta-vascular Case

Background:

Lung cancer is a greatest threat to people's health and life. CT image leads to unclear boundary segmentation. Segmentation of irregular nodules and complex structure, boundary information is not well considered and lung nodules have always been a hot topic.

Objective:

In this study, the pulmonary nodule segmentation is accomplished with the new graph cut algorithm. The problem of segmenting the juxta-pleural and juxta-vascular nodules was investigated which is based on graph cut algorithm.

Methods:

Firstly, the inflection points by the curvature was decided. Secondly, we used kernel graph cut to segment the nodules for the initial edge. Thirdly, the seeds points based on cast raying method is performed; lastly, a novel geodesic distance function is proposed to improve the graph cut algorithm and applied in lung nodules segmentation.

Results:

The new algorithm has been tested on total 258 nodules. Table 1 summarizes the morphologic features of all the nodules and given the results between the successful segmentation group and the poor/failed segmentation group. Figure 1 to Fig. (12) shows segmentation effect of Juxta-vascular nodules, Juxta-pleural nodules, and comparted with the other interactive segmentation methods.

Conclusion:

The experimental verification shows better results with our algorithm, the results will measure the volume numerical approach to nodule volume. The results of lung nodules segmentation in this study are as good as the results obtained by the other methods.

Multiple Resolution Residually Connected Feature Streams for Automatic Lung Tumor Segmentation From CT Images

Volumetric lung tumor segmentation and accurate longitudinal tracking of tumor volume changes from computed tomography images are essential for monitoring tumor response to therapy. Hence, we developed two multiple resolution residually connected network (MRRN) formulations called incremental-MRRN and dense-MRRN. Our networks simultaneously combine features across multiple image resolution and feature levels through residual connections to detect and segment the lung tumors. We evaluated our method on a total of 1210 non-small cell (NSCLC) lung tumors and nodules from three data sets consisting of 377 tumors from the open-source Cancer Imaging Archive (TCIA), 304 advanced stage NSCLC treated with anti- PD-1 checkpoint immunotherapy from internal institution MSKCC data set, and 529 lung nodules from the Lung Image Database Consortium (LIDC). The algorithm was trained using 377 tumors from the TCIA data set and validated on the MSKCC and tested on LIDC data sets. The segmentation accuracy compared to expert delineations was evaluated by computing the dice similarity coefficient, Hausdorff distances, sensitivity, and precision metrics. Our best performing incremental-MRRN method produced the highest DSC of 0.74 ± 0.13 for TCIA, 0.75±0.12 for MSKCC, and 0.68±0.23 for the LIDC data sets. There was no significant difference in the estimations of volumetric tumor changes computed using the incremental-MRRN method compared with the expert segmentation. In summary, we have developed a multi-scale CNN approach for volumetrically segmenting lung tumors which enables accurate, automated identification of and serial measurement of tumor volumes in the lung.

Cascaded classifiers and stacking methods for classification of pulmonary nodule characteristics

BACKGROUND AND OBJECTIVES

Detection and classification of pulmonary nodules are critical tasks in medical image analysis. The Lung Image Database Consortium (LIDC) database is a widely used resource for small pulmonary nodule classification research. This dataset is comprised of nodule characteristic evaluations and CT scans of patients. Although these characteristics are utilized in several studies, they can be used to improve classification performance.

METHODS

Numerous methods have been proposed to classify malignancy, but there are not many studies that facilitate nodule characteristics in classification steps. In this study, we use information on nodule characteristics and propose cascaded classification schemes. A group of hand-crafted features and deep features are used to define the nodules. In the first step of the classifier, the nodule characteristics are classified based on individual base classifiers. In the second step, the results of the first level classifier are combined for use in malignancy classification. In addition, stacking methods are applied to improve the performance of the cascaded classifiers.

RESULTS

The results confirmed that combining deep and hand-crafted features contribute to classification performance with an 8% improvement in average classification accuracy, 9% improvement in sensitivity, and 3% in specificity. Deep features from a nodule bounding area are more descriptive than the exact nodule region. The best performing cascaded classifier featured a classification accuracy of 84.70%, sensitivity of 67.37%, and specificity of 95.46%. First level stacking demonstrated similar results on classification accuracy and specificity but sensitivity was measured at 75.59%. Stacking on both levels provided the best classification accuracy and specificity with scores of 86.98% and 96.06%, respectively. When the malignancy ratings were grouped, stacking on both levels demonstrated better performance than other methods with a classification accuracy of 88.80%, sensitivity of 88.41%, and specificity of 94.12%.

CONCLUSIONS

Information on cascading characteristics with image features is beneficial for the classification of the malignancy ratings. Stacking approaches on both levels demonstrate better classification accuracy, but in the context of sensitivity, first level stacking performs better. Grouping the malignancy ratings results in better classification outcomes as in the case of similar studies in the literature.

Lung tumor segmentation methods: Impact on the uncertainty of radiomics features for non-small cell lung cancer

Purpose

To evaluate the uncertainty of radiomics features from contrast-enhanced breath-hold helical CT scans of non-small cell lung cancer for both manual and semi-automatic segmentation due to intra-observer, inter-observer, and inter-software reliability.

Methods

Three radiation oncologists manually delineated lung tumors twice from 10 CT scans using two software tools (3D-Slicer and MIM Maestro). Additionally, three observers without formal clinical training were instructed to use two semi-automatic segmentation tools, Lesion Sizing Toolkit (LSTK) and GrowCut, to delineate the same tumor volumes. The accuracy of the semi-automatic contours was assessed by comparison with physician manual contours using Dice similarity coefficients and Hausdorff distances. Eighty-three radiomics features were calculated for each delineated tumor contour. Informative features were identified based on their dynamic range and correlation to other features. Feature reliability was then evaluated using intra-class correlation coefficients (ICC). Feature range was used to evaluate the uncertainty of the segmentation methods.

Results

From the initial set of 83 features, 40 radiomics features were found to be informative, and these 40 features were used in the subsequent analyses. For both intra-observer and inter-observer reliability, LSTK had higher reliability than GrowCut and the two manual segmentation tools. All observers achieved consistently high ICC values when using LSTK, but the ICC value varied greatly for each observer when using GrowCut and the manual segmentation tools. For inter-software reliability, features were not reproducible across the software tools for either manual or semi-automatic segmentation methods. Additionally, no feature category was found to be more reproducible than another feature category. Feature ranges of LSTK contours were smaller than those of manual contours for all features.

Conclusion

Radiomics features extracted from LSTK contours were highly reliable across and among observers. With semi-automatic segmentation tools, observers without formal clinical training were comparable to physicians in evaluating tumor segmentation.

Automated glioma detection and segmentation using graphical models

Glioma detection and segmentation is a challenging task for radiologists and clinicians. The research reported in this paper seeks to develop a better clinical decision support algorithm for clinicians diagnosis. This paper presents a probabilistic method for detection and segmentation between abnormal tissue regions and brain tumour (tumour core and edema) portions from Magnetic Resonance Imaging (MRI). A framework is constructed to learn structure of undirected graphical models that can represent the spatial relationships among variables and apply it to glioma segmentation. Compared with the pixel of image, the superpixel is more consistent with human visual cognition and contains less redundancy, thus, the superpixels are considered as the basic unit of structure learning and glioma segmentation scheme. ℓ1-regularization techniques are applied to learn the appropriate structure for modeling graphical models. Conditional Random Fields (CRF) are used to model the spatial interactions among image superpixel regions and their measurements. A number of features including statistics features, the combined features from the local binary pattern as well as gray level run length, curve features, and fractal features were extracted from each superpixel. The features are then passed by ℓ1-regularization to ensure a robust classification. The proposed method is compared with support vector machine and Fuzzy c-means to classify each superpixel into normal and abnormal tissue. The proposed system is tested for the presence of low grade as well as high grade glioma tumors on images collected from BRATS2013, BRATS2015 data set and Henan Provincial People's Hospital (HNPPH) data set. The experiments performed provides similarity between segmented and truth image up to 91.5% by correlation method.

Development and Validation of Segmentation Method for Lung Cancer Volumetry on Chest CT

The set of criteria called Response Evaluation Criteria In Solid Tumors (RECIST) is used to evaluate the remedial effects of lung cancer, whereby the size of a lesion can be measured in one dimension (diameter). Volumetric evaluation is desirable for estimating the size of a lesion accurately, but there are several constraints and limitations to calculating the volume in clinical trials. In this study, we developed a method to detect lesions automatically, with minimal intervention by the user, and calculate their volume. Our proposed method, called a spherical region-growing method (SPRG), uses segmentation that starts from a seed point set by the user. SPRG is a modification of an existing region-growing method that is based on a sphere instead of pixels. The SPRG method detects lesions while preventing leakage to neighboring tissues, because the sphere is grown, i.e., neighboring voxels are added, only when all the voxels meet the required conditions. In this study, two radiologists segmented lung tumors using a manual method and the proposed method, and the results of both methods were compared. The proposed method showed a high sensitivity of 81.68-84.81% and a high dice similarity coefficient (DSC) of 0.86-0.88 compared with the manual method. In addition, the SPRG intraclass correlation coefficient (ICC) was 0.998 (CI 0.997-0.999, p < 0.01), showing that the SPRG method is highly reliable. If our proposed method is used for segmentation and volumetric measurement of lesions, then objective and accurate results and shorter data analysis time are possible.

Prediction of pathological nodal involvement by CT-based Radiomic features of the primary tumor in patients with clinically node-negative peripheral lung adenocarcinomas

PURPOSE

The purpose of this study was to investigate the potential of computed tomography (CT) based radiomic features of primary tumors to predict pathological nodal involvement in clinically node-negative (N0) peripheral lung adenocarcinomas.

METHODS

A total of 187 patients with clinical N0 peripheral lung adenocarcinomas who underwent preoperative CT scan and subsequently received systematic lymph node dissection were retrospectively reviewed. 219 quantitative 3D radiomic features of primary lung tumor were extracted; meanwhile, nine radiological semantic features were evaluated. Univariate and multivariate logistic regression analysis were used to explore the role of these features in predicting pathological nodal involvement. The areas under the ROC curves (AUCs) were compared between multivariate logistic regression models.

RESULTS

A total of 153 patients had pathological N0 status and 34 had pathological lymph node metastasis. On univariate analysis, fissure attachment and 17 radiomic features were significantly associated with pathological nodal involvement. Multivariate analysis revealed that semantic features of pleural retraction (P = 0.048) and fissure attachment (P = 0.023) were significant predictors of pathological nodal involvement (AUC = 0.659); and the radiomic feature F185 (Histogram SD Layer 1) (P = 0.0001) was an independent prognostic factor of pathological nodal involvement (AUC = 0.73). A logistic regression model produced from combining radiomic feature and semantic feature showed the highest AUC of 0.758 (95% CI: 0.685-0.831), and the AUC value computed by fivefold cross-validation method was 0.737 (95% CI: 0.73-0.744).

CONCLUSIONS

Features derived on primary lung tumor described by semantic and radiomic could provide information of pathological nodal involvement in clinical N0 peripheral lung adenocarcinomas.

Digital liver biopsy: Bio-imaging of fatty liver for translational and clinical research

The rapidly growing field of functional, molecular and structural bio-imaging is providing an extraordinary new opportunity to overcome the limits of invasive liver biopsy and introduce a "digital biopsy" for in vivo study of liver pathophysiology. To foster the application of bio-imaging in clinical and translational research, there is a need to standardize the methods of both acquisition and the storage of the bio-images of the liver. It can be hoped that the combination of digital, liquid and histologic liver biopsies will provide an innovative synergistic tri-dimensional approach to identifying new aetiologies, diagnostic and prognostic biomarkers and therapeutic targets for the optimization of personalized therapy of liver diseases and liver cancer. A group of experts of different disciplines (Special Interest Group for Personalized Hepatology of the Italian Association for the Study of the Liver, Institute for Biostructures and Bio-imaging of the National Research Council and Bio-banking and Biomolecular Resources Research Infrastructure) discussed criteria, methods and guidelines for facilitating the requisite application of data collection. This manuscript provides a multi-Author review of the issue with special focus on fatty liver.

Combining radiomic features with a miRNA classifier may improve prediction of malignant pathology for pancreatic intraductal papillary mucinous neoplasms

Intraductal papillary mucinous neoplasms (IPMNs) are pancreatic cancer precursors incidentally discovered by cross-sectional imaging. Consensus guidelines for IPMN management rely on standard radiologic features to predict pathology, but they lack accuracy. Using a retrospective cohort of 38 surgically-resected, pathologically-confirmed IPMNs (20 benign; 18 malignant) with preoperative computed tomography (CT) images and matched plasma-based ‘miRNA genomic classifier (MGC)’ data, we determined whether quantitative ‘radiomic’ CT features (+/- the MGC) can more accurately predict IPMN pathology than standard radiologic features ‘high-risk’ or ‘worrisome’ for malignancy. Logistic regression, principal component analyses, and cross-validation were used to examine associations. Sensitivity, specificity, positive and negative predictive value (PPV, NPV) were estimated. The MGC, ‘high-risk,’ and ‘worrisome’ radiologic features had area under the receiver operating characteristic curve (AUC) values of 0.83, 0.84, and 0.54, respectively. Fourteen radiomic features differentiated malignant from benign IPMNs (p<0.05) and collectively had an AUC=0.77. Combining radiomic features with the MGC revealed an AUC=0.92 and superior sensitivity (83%), specificity (89%), PPV (88%), and NPV (85%) than other models. Evaluation of uncertainty by 10-fold cross-validation retained an AUC>0.80 (0.87 (95% CI:0.84-0.89)). This proof-of-concept study suggests a noninvasive radiogenomic approach may more accurately predict IPMN pathology than ‘worrisome’ radiologic features considered in consensus guidelines.

A Comparison of Lung Nodule Segmentation Algorithms: Methods and Results from a Multi-institutional Study

Tumor volume estimation, as well as accurate and reproducible borders segmentation in medical images, are important in the diagnosis, staging, and assessment of response to cancer therapy. The goal of this study was to demonstrate the feasibility of a multi-institutional effort to assess the repeatability and reproducibility of nodule borders and volume estimate bias of computerized segmentation algorithms in CT images of lung cancer, and to provide results from such a study. The dataset used for this evaluation consisted of 52 tumors in 41 CT volumes (40 patient datasets and 1 dataset containing scans of 12 phantom nodules of known volume) from five collections available in The Cancer Imaging Archive. Three academic institutions developing lung nodule segmentation algorithms submitted results for three repeat runs for each of the nodules. We compared the performance of lung nodule segmentation algorithms by assessing several measurements of spatial overlap and volume measurement. Nodule sizes varied from 29 μl to 66 ml and demonstrated a diversity of shapes. Agreement in spatial overlap of segmentations was significantly higher for multiple runs of the same algorithm than between segmentations generated by different algorithms (p < 0.05) and was significantly higher on the phantom dataset compared to the other datasets (p < 0.05). Algorithms differed significantly in the bias of the measured volumes of the phantom nodules (p < 0.05) underscoring the need for assessing performance on clinical data in addition to phantoms. Algorithms that most accurately estimated nodule volumes were not the most repeatable, emphasizing the need to evaluate both their accuracy and precision. There were considerable differences between algorithms, especially in a subset of heterogeneous nodules, underscoring the recommendation that the same software be used at all time points in longitudinal studies.

Semi-automated pulmonary nodule interval segmentation using the NLST data

PURPOSE

To study the variability in volume change estimates of pulmonary nodules due to segmentation approaches used across several algorithms and to evaluate these effects on the ability to predict nodule malignancy.

METHODS

We obtained 100 patient image datasets from the National Lung Screening Trial (NLST) that had a nodule detected on each of two consecutive low dose computed tomography (LDCT) scans, with an equal proportion of malignant and benign cases (50 malignant, 50 benign). Information about the nodule location for the cases was provided by a screen capture with a bounding box and its axial location was indicated. Five participating quantitative imaging network (QIN) institutions performed nodule segmentation using their preferred semi-automated algorithms with no manual correction; teams were allowed to provide additional manually corrected segmentations (analyzed separately). The teams were asked to provide segmentation masks for each nodule at both time points. From these masks, the volume was estimated for the nodule at each time point; the change in volume (absolute and percent change) across time points was estimated as well. We used the concordance correlation coefficient (CCC) to compare the similarity of computed nodule volumes (absolute and percent change) across algorithms. We used Logistic regression model on the change in volume (absolute change and percent change) of the nodules to predict the malignancy status, the area under the receiver operating characteristic curve (AUROC) and confidence intervals were reported. Because the size of nodules was expected to have a substantial effect on segmentation variability, analysis of change in volumes was stratified by lesion size, where lesions were grouped into those with a longest diameter of <8 mm and those with longest diameter ≥ 8 mm.

RESULTS

We find that segmentation of the nodules shows substantial variability across algorithms, with the CCC ranging from 0.56 to 0.95 for change in volume (percent change in volume range was [0.15 to 0.86]) across the nodules. When examining nodules based on their longest diameter, we find the CCC had higher values for large nodules with a range of [0.54 to 0.93] among the algorithms, while percent change in volume was [0.3 to 0.95]. Compared to that of smaller nodules which had a range of [-0.0038 to 0.69] and percent change in volume was [-0.039 to 0.92]. The malignancy prediction results showed fairly consistent results across the institutions, the AUC using change in volume ranged from 0.65 to 0.89 (Percent change in volume was 0.64 to 0.86) for entire nodule range. Prediction improves for large nodule range (≥ 8 mm) with AUC range 0.75 to 0.90 (percent change in volume was 0.74 to 0.92). Compared to smaller nodule range (<8 mm) with AUC range 0.57 to 0.78 (percent change in volume was 0.59 to 0.77).

CONCLUSIONS

We find there is a fairly high concordance in the size measurements for larger nodules (≥8 mm) than the lower sizes (<8 mm) across algorithms. We find the change in nodule volume (absolute and percent change) were consistent predictors of malignancy across institutions, despite using different segmentation algorithms. Using volume change estimates without corrections shows slightly lower predictability (for two teams).

Added Value of Computer-aided CT Image Features for Early Lung Cancer Diagnosis with Small Pulmonary Nodules: A Matched Case-Control Study

Purpose To test whether computer-aided diagnosis (CAD) approaches can increase the positive predictive value (PPV) and reduce the false-positive rate in lung cancer screening for small nodules compared with human reading by thoracic radiologists. Materials and Methods A matched case-control sample of low-dose computed tomography (CT) studies in 186 participants with 4-20-mm noncalcified lung nodules who underwent biopsy in the National Lung Screening Trial (NLST) was selected. Variables used for matching were age, sex, smoking status, chronic obstructive pulmonary disease status, body mass index, study year of the positive screening test, and screening results. Studies before lung biopsy were randomly split into a training set (70 cancers plus 70 benign controls) and a validation set (20 cancers plus 26 benign controls). Image features from within and outside dominant nodules were extracted. A CAD algorithm developed from the training set and a random forest classifier were applied to the validation set to predict biopsy outcomes. Receiver operating characteristic analysis was used to compare the prediction accuracy of CAD with the NLST investigator's diagnosis and readings from three experienced and board-certified thoracic radiologists who used contemporary clinical practice guidelines. Results In the validation cohort, the area under the receiver operating characteristic curve for CAD was 0.9154. By default, the sensitivity, specificity, and PPV of the NLST investigators were 1.00, 0.00, and 0.43, respectively. The sensitivity, specificity, PPV, and negative predictive value of CAD and the three radiologists' combined reading were 0.95, 0.88, 0.86, and 0.96 and 0.70, 0.69, 0.64, and 0.75, respectively. Conclusion CAD could increase PPV and reduce the false-positive rate in the early diagnosis of lung cancer. © RSNA, 2017 Online supplemental material is available for this article.

Radiomics for Response and Outcome Assessment for Non-Small Cell Lung Cancer

Routine follow-up visits and radiographic imaging are required for outcome evaluation and tumor recurrence monitoring. Yet more personalized surveillance is required in order to sufficiently address the nature of heterogeneity in nonsmall cell lung cancer and possible recurrences upon completion of treatment. Radiomics, an emerging noninvasive technology using medical imaging analysis and data mining methodology, has been adopted to the area of cancer diagnostics in recent years. Its potential application in response assessment for cancer treatment has also drawn considerable attention. Radiomics seeks to extract a large amount of valuable information from patients' medical images (both pretreatment and follow-up images) and quantitatively correlate image features with diagnostic and therapeutic outcomes. Radiomics relies on computers to identify and analyze vast amounts of quantitative image features that were previously overlooked, unmanageable, or failed to be identified (and recorded) by human eyes. The research area has been focusing on the predictive accuracy of pretreatment features for outcome and response and the early discovery of signs of tumor response, recurrence, distant metastasis, radiation-induced lung injury, death, and other outcomes, respectively. This review summarized the application of radiomics in response assessments in radiotherapy and chemotherapy for non-small cell lung cancer, including image acquisition/reconstruction, region of interest definition/segmentation, feature extraction, and feature selection and classification. The literature search for references of this article includes PubMed peer-reviewed publications over the last 10 years on the topics of radiomics, textural features, radiotherapy, chemotherapy, lung cancer, and response assessment. Summary tables of radiomics in response assessment and treatment outcome prediction in radiation oncology have been developed based on the comprehensive review of the literature.

Radial gradient and radial deviation radiomic features from pre-surgical CT scans are associated with survival among lung adenocarcinoma patients

The goal of this study was to extract features from radial deviation and radial gradient maps which were derived from thoracic CT scans of patients diagnosed with lung adenocarcinoma and assess whether these features are associated with overall survival. We used two independent cohorts from different institutions for training (n= 61) and test (n= 47) and focused our analyses on features that were non-redundant and highly reproducible. To reduce the number of features and covariates into a single parsimonious model, a backward elimination approach was applied. Out of 48 features that were extracted, 31 were eliminated because they were not reproducible or were redundant. We considered 17 features for statistical analysis and identified a final model containing the two most highly informative features that were associated with lung cancer survival. One of the two features, radial deviation outside-border separation standard deviation, was replicated in a test cohort exhibiting a statistically significant association with lung cancer survival (multivariable hazard ratio = 0.40; 95% confidence interval 0.17-0.97). Additionally, we explored the biological underpinnings of these features and found radial gradient and radial deviation image features were significantly associated with semantic radiological features.

CT imaging features associated with recurrence in non-small cell lung cancer patients after stereotactic body radiotherapy

BACKGROUND

Predicting recurrence after stereotactic body radiotherapy (SBRT) in non-small cell lung cancer (NSCLC) patients is problematic, but critical for the decision of following treatment. This study aims to investigate the association of imaging features derived from the first follow-up computed tomography (CT) on lung cancer patient outcomes following SBRT, and identify patients at high risk of recurrence.

METHODS

Fifty nine biopsy-proven non-small cell lung cancer patients were qualified for this study. The first follow-up CTs were performed about 3 months after SBRT (median time: 91 days). Imaging features included 34 manually scored radiological features (semantics) describing the lesion, lung and thorax and 219 quantitative imaging features (radiomics) extracted automatically after delineation of the lesion. Cox proportional hazard models and Harrel's C-index were used to explore predictors of overall survival (OS), recurrence-free survival (RFS), and loco-regional recurrence-free survival (LR-RFS). Five-fold cross validation was performed on the final prognostic model.

RESULTS

The median follow-up time was 42 months. The model for OS contained Eastern Cooperative Oncology Group (ECOG) performance status (HR = 3.13, 95% CI: 1.17-8.41), vascular involvement (HR = 3.21, 95% CI: 1.29-8.03), lymphadenopathy (HR = 3.59, 95% CI: 1.58-8.16) and the 1st principle component of radiomic features (HR = 1.24, 95% CI: 1.02-1.51). The model for RFS contained vascular involvement (HR = 3.06, 95% CI: 1.40-6.70), vessel attachment (HR = 3.46, 95% CI: 1.65-7.25), pleural retraction (HR = 3.24, 95% CI: 1.41-7.42), lymphadenopathy (HR = 6.41, 95% CI: 2.58-15.90) and relative enhancement (HR = 1.40, 95% CI: 1.00-1.96). The model for LR-RFS contained vascular involvement (HR = 4.96, 95% CI: 2.23-11.03), lymphadenopathy (HR = 2.64, 95% CI: 1.19-5.82), circularity (F13, HR = 1.60, 95% CI: 1.10-2.32) and 3D Laws feature (F92, HR = 1.96, 95% CI: 1.35-2.83). Five-fold cross-validated the areas under the receiver operating characteristic curves (AUC) of these three models were all above 0.8.

CONCLUSIONS

Our analysis reveals disease progression could be prognosticated as early as 3 months after SBRT using CT imaging features, and these features would be helpful in clinical decision-making.

Semiautomatic Segmentation of Glioma on Mobile Devices

Brain tumor segmentation is the first and the most critical step in clinical applications of radiomics. However, segmenting brain images by radiologists is labor intense and prone to inter- and intraobserver variability. Stable and reproducible brain image segmentation algorithms are thus important for successful tumor detection in radiomics. In this paper, we propose a supervised brain image segmentation method, especially for magnetic resonance (MR) brain images with glioma. This paper uses hard edge multiplicative intrinsic component optimization to preprocess glioma medical image on the server side, and then, the doctors could supervise the segmentation process on mobile devices in their convenient time. Since the preprocessed images have the same brightness for the same tissue voxels, they have small data size (typically 1/10 of the original image size) and simple structure of 4 types of intensity value. This observation thus allows follow-up steps to be processed on mobile devices with low bandwidth and limited computing performance. Experiments conducted on 1935 brain slices from 129 patients show that more than 30% of the sample can reach 90% similarity; over 60% of the samples can reach 85% similarity, and more than 80% of the sample could reach 75% similarity. The comparisons with other segmentation methods also demonstrate both efficiency and stability of the proposed approach.

Imaging features from pretreatment CT scans are associated with clinical outcomes in nonsmall-cell lung cancer patients treated with stereotactic body radiotherapy

PURPOSE

To investigate whether imaging features from pretreatment planning CT scans are associated with overall survival (OS), recurrence-free survival (RFS), and loco-regional recurrence-free survival (LR-RFS) after stereotactic body radiotherapy (SBRT) among nonsmall-cell lung cancer (NSCLC) patients.

PATIENTS AND METHODS

A total of 92 patients (median age: 73 yr) with stage I or IIA NSCLC were qualified for this study. A total dose of 50 Gy in five fractions was the standard treatment. Besides clinical characteristics, 24 "semantic" image features were manually scored based on a point scale (up to 5) and 219 computer-derived "radiomic" features were extracted based on whole tumor segmentation. Statistical analysis was performed using Cox proportional hazards model and Harrell's C-index, and the robustness of final prognostic model was assessed using tenfold cross validation by dichotomizing patients according to the survival or recurrence status at 24 months.

RESULTS

Two-year OS, RFS and LR-RFS were 69.95%, 41.3%, and 51.85%, respectively. There was an improvement of Harrell's C-index when adding imaging features to a clinical model. The model for OS contained the Eastern Cooperative Oncology Group (ECOG) performance status [Hazard Ratio (HR) = 2.78, 95% Confidence Interval (CI): 1.37-5.65], pleural retraction (HR = 0.27, 95% CI: 0.08-0.92), F2 (short axis × longest diameter, HR = 1.72, 95% CI: 1.21-2.44) and F186 (Hist-Energy-L1, HR = 1.27, 95% CI: 1.00-1.61); The prognostic model for RFS contained vessel attachment (HR = 2.13, 95% CI: 1.24-3.64) and F2 (HR = 1.69, 95% CI: 1.33-2.15); and the model for LR-RFS contained the ECOG performance status (HR = 2.01, 95% CI: 1.12-3.60) and F2 (HR = 1.67, 95% CI: 1.29-2.18).

CONCLUSIONS

Imaging features derived from planning CT demonstrate prognostic value for recurrence following SBRT treatment, and might be helpful in patient stratification.

Multi-phase simultaneous segmentation of tumor in lung 4D-CT data with context information

Lung 4D computed tomography (4D-CT) plays an important role in high-precision radiotherapy because it characterizes respiratory motion, which is crucial for accurate target definition. However, the manual segmentation of a lung tumor is a heavy workload for doctors because of the large number of lung 4D-CT data slices. Meanwhile, tumor segmentation is still a notoriously challenging problem in computer-aided diagnosis. In this paper, we propose a new method based on an improved graph cut algorithm with context information constraint to find a convenient and robust approach of lung 4D-CT tumor segmentation. We combine all phases of the lung 4D-CT into a global graph, and construct a global energy function accordingly. The sub-graph is first constructed for each phase. A context cost term is enforced to achieve segmentation results in every phase by adding a context constraint between neighboring phases. A global energy function is finally constructed by combining all cost terms. The optimization is achieved by solving a max-flow/min-cut problem, which leads to simultaneous and robust segmentation of the tumor in all the lung 4D-CT phases. The effectiveness of our approach is validated through experiments on 10 different lung 4D-CT cases. The comparison with the graph cut without context constraint, the level set method and the graph cut with star shape prior demonstrates that the proposed method obtains more accurate and robust segmentation results.

Associations between radiologist-defined semantic and automatically computed radiomic features in non-small cell lung cancer

Tumor phenotypes captured in computed tomography (CT) images can be described qualitatively and quantitatively using radiologist-defined "semantic" and computer-derived "radiomic" features, respectively. While both types of features have shown to be promising predictors of prognosis, the association between these groups of features remains unclear. We investigated the associations between semantic and radiomic features in CT images of 258 non-small cell lung adenocarcinomas. The tumor imaging phenotypes were described using 9 qualitative semantic features that were scored by radiologists, and 57 quantitative radiomic features that were automatically calculated using mathematical algorithms. Of the 9 semantic features, 3 were rated on a binary scale (cavitation, air bronchogram, and calcification) and 6 were rated on a categorical scale (texture, border definition, contour, lobulation, spiculation, and concavity). 32-41 radiomic features were associated with the binary semantic features (AUC = 0.56-0.76). The relationship between all radiomic features and the categorical semantic features ranged from weak to moderate (|Spearmen's correlation| = 0.002-0.65). There are associations between semantic and radiomic features, however the associations were not strong despite being significant. Our results indicate that radiomic features may capture distinct tumor phenotypes that fail to be perceived by naked eye that semantic features do not describe and vice versa.

Quantitative Clinical Imaging Methods for Monitoring Intratumoral Evolution

Solid tumors are multiscale, open, complex, dynamic systems: complex because they have many interacting components, dynamic because both the components and their interactions can change with time, and open because the tumor freely communicates with surrounding and even distant host tissue. Thus, it is not surprising that striking intratumoral variations are commonly observed in clinical imaging such as MRI and CT and that several recent studies found striking regional variations in the molecular properties of cancer cells from the same tumor. Interestingly, this spatial heterogeneity in molecular properties of tumor cells is typically ascribed to branching clonal evolution due to accumulating mutations while macroscopic variations observed in, for example, clinical MRI scans are usually viewed as functions of blood flow. The clinical significance of spatial heterogeneity has not been fully determined but there is a general consensus that the varying intratumoral landscape along with patient factors such as age, morbidity and lifestyle, contributes significantly to the often unpredictable response of individual patients within a disease cohort treated with the same standard-of-care therapy.Here we investigate the potential link between macroscopic tumor heterogeneity observed by clinical imaging and spatial variations in the observed molecular properties of cancer cells. We build on techniques developed in landscape ecology to link regional variations in the distribution of species with local environmental conditions that define their habitat. That is, we view each region of the tumor as a local ecosystem consisting of environmental conditions such as access to nutrients, oxygen, and means of waste clearance related to blood flow and the local population of tumor cells that both adapt to these conditions and, to some extent, change them through, for example, production of angiogenic factors. Furthermore, interactions among neighboring habitats can produce broader regional dynamics so that the internal diversity of tumors is the net result of complex multiscale somatic Darwinian interactions.Methods in landscape ecology harness Darwinian dynamics to link the environmental properties of a given region to the local populations which are assumed to represent maximally fit phenotypes within those conditions. Consider a common task of a landscape ecologist: defining the spatial distribution of species in a large region, e.g., in a satellite image. Clearly the most accurate approach requires a meter by meter survey of the multiple square kilometers in the region of interest. However, this is both impractical and potentially destructive. Instead, landscape ecology breaks the task into component parts relying on the Darwinian interdependence of environmental properties and fitness of specific species' phenotypic and genotypic properties. First, the satellite map is carefully analyzed to define the number and distribution of habitats. Then the species distribution in a representative sampling of each habitat is empirically determined. Ultimately, this permits sufficient bridging of spatial scales to accurately predict spatial distribution of plant and animal species within large regions.Currently, identifying intratumoral subpopulations requires detailed histological and molecular studies that are expensive and time consuming. Furthermore, this method is subject to sampling bias, is invasive for vital organs such as the brain, and inherently destructive precluding repeated assessments for monitoring post-treatment response and proteogenomic evolution. In contrast, modern cross-sectional imaging can interrogate the entire tumor noninvasively, allowing repeated analysis without disrupting the region of interest. In particular, magnetic resonance imaging (MRI) provides exceptional spatial resolution and generates signals that are unique to the molecular constituents of tissue. Here we propose that MRI scans may be the equivalent of satellite images in landscape ecology and, with appropriate application of Darwinian first principles and sophisticated image analytic methods, can be used to estimate regional variations in the molecular properties of cancer cells.We have initially examined this technique in glioblastoma, a malignant brain neoplasm which is morphologically complex and notorious for a fast progression from diagnosis to recurrence and death, making a suitable subject of noninvasive, rapidly repeated assessment of intratumoral evolution. Quantitative imaging analysis of routine clinical MRIs from glioblastoma has identified macroscopic morphologic characteristics which correlate with proteogenomics and prognosis. The key to the accurate detection and forecasting of intratumoral evolution using quantitative imaging analysis is likely to be in the understanding of the synergistic interactions between observable intratumoral subregions and the resulting tumor behavior.

Deep Feature Transfer Learning in Combination with Traditional Features Predicts Survival Among Patients with Lung Adenocarcinoma

Lung cancer is the most common cause of cancer-related deaths in the USA. It can be detected and diagnosed using computed tomography images. For an automated classifier, identifying predictive features from medical images is a key concern. Deep feature extraction using pretrained convolutional neural networks (CNNs) has recently been successfully applied in some image domains. Here, we applied a pretrained CNN to extract deep features from 40 computed tomography images, with contrast, of non-small cell adenocarcinoma lung cancer, and combined deep features with traditional image features and trained classifiers to predict short- and long-term survivors. We experimented with several pretrained CNNs and several feature selection strategies. The best previously reported accuracy when using traditional quantitative features was 77.5% (area under the curve [AUC], 0.712), which was achieved by a decision tree classifier. The best reported accuracy from transfer learning and deep features was 77.5% (AUC, 0.713) using a decision tree classifier. When extracted deep neural network features were combined with traditional quantitative features, we obtained an accuracy of 90% (AUC, 0.935) with the 5 best post-rectified linear unit features extracted from a vgg-f pretrained CNN and the 5 best traditional features. The best results were achieved with the symmetric uncertainty feature ranking algorithm followed by a random forests classifier.

A Rapid Segmentation-Insensitive "Digital Biopsy" Method for Radiomic Feature Extraction: Method and Pilot Study Using CT Images of Non-Small Cell Lung Cancer

Quantitative imaging approaches compute features within images' regions of interest. Segmentation is rarely completely automatic, requiring time-consuming editing by experts. We propose a new paradigm, called “digital biopsy,” that allows for the collection of intensity- and texture-based features from these regions at least 1 order of magnitude faster than the current manual or semiautomated methods. A radiologist reviewed automated segmentations of lung nodules from 100 preoperative volume computed tomography scans of patients with non–small cell lung cancer, and manually adjusted the nodule boundaries in each section, to be used as a reference standard, requiring up to 45 minutes per nodule. We also asked a different expert to generate a digital biopsy for each patient using a paintbrush tool to paint a contiguous region of each tumor over multiple cross-sections, a procedure that required an average of <3 minutes per nodule. We simulated additional digital biopsies using morphological procedures. Finally, we compared the features extracted from these digital biopsies with our reference standard using intraclass correlation coefficient (ICC) to characterize robustness. Comparing the reference standard segmentations to our digital biopsies, we found that 84/94 features had an ICC >0.7; comparing erosions and dilations, using a sphere of 1.5-mm radius, of our digital biopsies to the reference standard segmentations resulted in 41/94 and 53/94 features, respectively, with ICCs >0.7. We conclude that many intensity- and texture-based features remain consistent between the reference standard and our method while substantially reducing the amount of operator time required.

Predicting Malignant Nodules from Screening CT Scans

OBJECTIVES

The aim of this study was to determine whether quantitative analyses ("radiomics") of low-dose computed tomography lung cancer screening images at baseline can predict subsequent emergence of cancer.

METHODS

Public data from the National Lung Screening Trial (ACRIN 6684) were assembled into two cohorts of 104 and 92 patients with screen-detected lung cancer and then matched with cohorts of 208 and 196 screening subjects with benign pulmonary nodules. Image features were extracted from each nodule and used to predict the subsequent emergence of cancer.

RESULTS

The best models used 23 stable features in a random forests classifier and could predict nodules that would become cancerous 1 and 2 years hence with accuracies of 80% (area under the curve 0.83) and 79% (area under the curve 0.75), respectively. Radiomics outperformed the Lung Imaging Reporting and Data System and volume-only approaches. The performance of the McWilliams risk assessment model was commensurate.

CONCLUSIONS

The radiomics of lung cancer screening computed tomography scans at baseline can be used to assess risk for development of cancer.