Lukas Ebner

Lung Pattern Classification for Interstitial Lung Diseases Using a Deep Convolutional Neural Network

Automated tissue characterization is one of the most crucial components of a computer aided diagnosis (CAD) system for interstitial lung diseases (ILDs). Although much research has been conducted in this field, the problem remains challenging. Deep learning techniques have recently achieved impressive results in a variety of computer vision problems, raising expectations that they might be applied in other domains, such as medical image analysis. In this paper, we propose and evaluate a convolutional neural network (CNN), designed for the classification of ILD patterns. The proposed network consists of 5 convolutional layers with 2 × 2 kernels and LeakyReLU activations, followed by average pooling with size equal to the size of the final feature maps and three dense layers. The last dense layer has 7 outputs, equivalent to the classes considered: healthy, ground glass opacity (GGO), micronodules, consolidation, reticulation, honeycombing and a combination of GGO/reticulation. To train and evaluate the CNN, we used a dataset of 14696 image patches, derived by 120 CT scans from different scanners and hospitals. To the best of our knowledge, this is the first deep CNN designed for the specific problem. A comparative analysis proved the effectiveness of the proposed CNN against previous methods in a challenging dataset. The classification performance ( ~ 85.5%) demonstrated the potential of CNNs in analyzing lung patterns. Future work includes, extending the CNN to three-dimensional data provided by CT volume scans and integrating the proposed method into a CAD system that aims to provide differential diagnosis for ILDs as a supportive tool for radiologists.

Postmortem whole-body MRI in traumatic causes of death

OBJECTIVE The aim of this study was to determine the sensitivity and specificity of postmortem whole-body MRI for typical injuries resulting from traumatic causes of death. MATERIALS AND METHODS Forty cases of accidental death were evaluated with postmortem whole-body MRI. Imaging was conducted according to a standard protocol, and each examination had an average duration of 90 minutes. The imaging findings were correlated with the autopsy findings, which served as the reference standard. RESULTS MRI showed the main pathologic process leading to death in 39 of the 40 cases. The sensitivity of postmortem MRI ranged from 100% (pneumothorax) to 40% (fractures of the upper extremities). In general, MRI had a high level of performance for depicting soft-tissue lesions, such as subcutaneous hematoma (e.g., galeal hematoma with a sensitivity 95%). The sensitivity of MRI was remarkably lower for lesions of the upper abdominal organs (liver, 80%; spleen, 50%; pancreas, 60%; kidneys, 66%). CONCLUSION Postmortem whole-body MRI had overall good performance for depicting traumatic findings in corpses and therefore may serve an important role as an adjunct to classic autopsy for the forensic examination of cases of traumatic cause of death. However, the reduced sensitivity of postmortem MRI for lacerations of the upper abdominal organs and the observed superimposition of antemortem findings and postmortem findings (e.g., in the pulmonary tissue) in this retrospective study suggest that whole-body postmortem MRI not be recommended as a replacement for classic autopsy.

CT dose and image quality in the last three scanner generations

AIM To compare the computed tomography (CT) dose and image quality with the filtered back projection against the iterative reconstruction and CT with a minimal electronic noise detector. METHODS A lung phantom (Chest Phantom N1 by Kyoto Kagaku) was scanned with 3 different CT scanners: the Somatom Sensation, the Definition Flash and the Definition Edge (all from Siemens, Erlangen, Germany). The scan parameters were identical to the Siemens presetting for THORAX ROUTINE (scan length 35 cm and FOV 33 cm). Nine different exposition levels were examined (reference mAs/peek voltage): 100/120, 100/100, 100/80, 50/120, 50/100, 50/80, 25/120, 25/100 and 25 mAs/80 kVp. Images from the SOMATOM Sensation were reconstructed using classic filtered back projection. Iterative reconstruction (SAFIRE, level 3) was performed for the two other scanners. A Stellar detector was used with the Somatom Definition Edge. The CT doses were represented by the dose length products (DLPs) (mGycm) provided by the scanners. Signal, contrast, noise and subjective image quality were recorded by two different radiologists with 10 and 3 years of experience in chest CT radiology. To determine the average dose reduction between two scanners, the integral of the dose difference was calculated from the lowest to the highest noise level. RESULTS When using iterative reconstruction (IR) instead of filtered back projection (FBP), the average dose reduction was 30%, 52% and 80% for bone, soft tissue and air, respectively, for the same image quality (P < 0.0001). The recently introduced Stellar detector (Sd) lowered the radiation dose by an additional 27%, 54% and 70% for bone, soft tissue and air, respectively (P < 0.0001). The benefit of dose reduction was larger at lower dose levels. With the same radiation dose, an average of 34% (22%-37%) and 25% (13%-46%) more contrast to noise was achieved by changing from FBP to IR and from IR to Sd, respectively. For the same contrast to noise level, an average of 59% (46%-71%) and 51% (38%-68%) dose reduction was produced for IR and Sd, respectively. For the same subjective image quality, the dose could be reduced by 25% (2%-42%) and 44% (33%-54%) using IR and Sd, respectively. CONCLUSION This study showed an average dose reduction between 27% and 70% for the new Stellar detector, which is equivalent to using IR instead of FBP.

Lung cancer screening with CT: Evaluation of radiologists and different computer assisted detection software (CAD) as first and second readers for lung nodule detection at different dose levels

OBJECTIVES To find the best pairing of first and second reader at highest sensitivity for detecting lung nodules with CT at various dose levels. MATERIALS AND METHODS An anthropomorphic lung phantom and artificial lung nodules were used to simulate screening CT-examination at standard dose (100 mAs, 120 kVp) and 8 different low dose levels, using 120, 100 and 80 kVp combined with 100, 50 and 25 mAs. At each dose level 40 phantoms were randomly filled with 75 solid and 25 ground glass nodules (5-12 mm). Two radiologists and 3 different computer aided detection softwares (CAD) were paired to find the highest sensitivity. RESULTS Sensitivities at standard dose were 92%, 90%, 84%, 79% and 73% for reader 1, 2, CAD1, CAD2, CAD3, respectively. Combined sensitivity for human readers 1 and 2 improved to 97%, (p1=0.063, p2=0.016). Highest sensitivities--between 97% and 99.0%--were achieved by combining any radiologist with any CAD at any dose level. Combining any two CADs, sensitivities between 85% and 88% were significantly lower than for radiologists combined with CAD (p<0.03). CONCLUSIONS Combination of a human observer with any of the tested CAD systems provide optimal sensitivity for lung nodule detection even at reduced dose at 25 mAs/80 kVp.

Comparison of Dual-Energy Subtraction and Electronic Bone Suppression Combined With Computer-Aided Detection on Chest Radiographs: Effect on Human Observers' Performance in Nodule Detection

OBJECTIVE The objective of our study was to compare the effect of dual-energy subtraction and bone suppression software alone and in combination with computer-aided detection (CAD) on the performance of human observers in lung nodule detection. MATERIALS AND METHODS One hundred one patients with from one to five lung nodules measuring 5-29 mm and 42 subjects with no nodules were retrospectively selected and randomized. Three independent radiologists marked suspicious-appearing lesions on the original chest radiographs, dual-energy subtraction images, and bone-suppressed images before and after postprocessing with CAD. Marks of the observers and CAD marks were compared with CT as the reference standard. Data were analyzed using nonparametric tests and the jackknife alternative free-response receiver operating characteristic (JAFROC) method. RESULTS Using dual-energy subtraction alone (p = 0.0198) or CAD alone (p = 0.0095) improved the detection rate compared with using the original conventional chest radiograph. The combination of bone suppression and CAD provided the highest sensitivity (51.6%) and the original nonenhanced conventional chest radiograph alone provided the lowest (46.9%; p = 0.0049). Dual-energy subtraction and bone suppression provided the same false-positive (p = 0.2702) and true-positive (p = 0.8451) rates. Up to 22.9% of lesions were found only by the CAD program and were missed by the readers. JAFROC showed no difference in the performance between modalities (p = 0.2742-0.5442). CONCLUSION Dual-energy subtraction and the electronic bone suppression program used in this study provided similar detection rates for pulmonary nodules. Additionally, CAD alone or combined with bone suppression can significantly improve the sensitivity of human observers for pulmonary nodule detection.

Multi-source Transfer Learning with Convolutional Neural Networks for Lung Pattern Analysis

Early diagnosis of interstitial lung diseases is crucial for their treatment, but even experienced physicians find it difficult, as their clinical manifestations are similar. In order to assist with the diagnosis, computer-aided diagnosis systems have been developed. These commonly rely on a fixed scale classifier that scans CT images, recognizes textural lung patterns, and generates a map of pathologies. In a previous study, we proposed a method for classifying lung tissue patterns using a deep convolutional neural network (CNN), with an architecture designed for the specific problem. In this study, we present an improved method for training the proposed network by transferring knowledge from the similar domain of general texture classification. Six publicly available texture databases are used to pretrain networks with the proposed architecture, which are then fine-tuned on the lung tissue data. The resulting CNNs are combined in an ensemble and their fused knowledge is compressed back to a network with the original architecture. The proposed approach resulted in an absolute increase of about 2% in the performance of the proposed CNN. The results demonstrate the potential of transfer learning in the field of medical image analysis, indicate the textural nature of the problem and show that the method used for training a network can be as important as designing its architecture.

Feasible Dose Reduction in Routine Chest Computed Tomography Maintaining Constant Image Quality Using the Last Three Scanner Generations: From Filtered Back Projection to Sinogram-affirmed Iterative Reconstruction and Impact of the Novel Fully Integrated Detector Design Minimizing Electronic Noise

Objective: The aim of the present study was to evaluate a dose reduction in contrast-enhanced chest computed tomography (CT) by comparing the three latest generations of Siemens CT scanners used in clinical practice. We analyzed the amount of radiation used with filtered back projection (FBP) and an iterative reconstruction (IR) algorithm to yield the same image quality. Furthermore, the influence on the radiation dose of the most recent integrated circuit detector (ICD; Stellar detector, Siemens Healthcare, Erlangen, Germany) was investigated. Materials and Methods: 136 Patients were included. Scan parameters were set to a thorax routine: SOMATOM Sensation 64 (FBP), SOMATOM Definition Flash (IR), and SOMATOM Definition Edge (ICD and IR). Tube current was set constantly to the reference level of 100 mA automated tube current modulation using reference milliamperes. Care kV was used on the Flash and Edge scanner, while tube potential was individually selected between 100 and 140 kVp by the medical technologists at the SOMATOM Sensation. Quality assessment was performed on soft-tissue kernel reconstruction. Dose was represented by the dose length product. Results: Dose-length product (DLP) with FBP for the average chest CT was 308 mGy*cm ± 99.6. In contrast, the DLP for the chest CT with IR algorithm was 196.8 mGy*cm ± 68.8 (P = 0.0001). Further decline in dose can be noted with IR and the ICD: DLP: 166.4 mGy*cm ± 54.5 (P = 0.033). The dose reduction compared to FBP was 36.1% with IR and 45.6% with IR/ICD. Signal-to-noise ratio (SNR) was favorable in the aorta, bone, and soft tissue for IR/ICD in combination compared to FBP (the P values ranged from 0.003 to 0.048). Overall contrast-to-noise ratio (CNR) improved with declining DLP. Conclusion: The most recent technical developments, namely IR in combination with integrated circuit detectors, can significantly lower radiation dose in chest CT examinations.

Using hyperpolarized 129Xe MRI to quantify regional gas transfer in idiopathic pulmonary fibrosis

Background Assessing functional impairment, therapeutic response and disease progression in patients with idiopathic pulmonary fibrosis (IPF) continues to be challenging. Hyperpolarized 129Xe MRI can address this gap through its unique capability to image gas transfer three-dimensionally from airspaces to interstitial barrier tissues to red blood cells (RBCs). This must be validated by testing the degree to which it correlates with pulmonary function tests (PFTs) and CT scores, and its spatial distribution reflects known physiology and patterns of disease. Methods 13 healthy individuals (33.6±15.7 years) and 12 patients with IPF (66.0±6.4 years) underwent 129Xe MRI to generate three-dimensional quantitative maps depicting the 129Xe ventilation distribution, its uptake in interstitial barrier tissues and its transfer to RBCs. For each map, mean values were correlated with PFTs and CT fibrosis scores, and their patterns were tested for the ability to depict functional gravitational gradients in healthy lung and to detect the known basal and peripheral predominance of disease in IPF. Results 129Xe MRI depicted functional impairment in patients with IPF, whose mean barrier uptake increased by 188% compared with the healthy reference population. 129Xe MRI metrics correlated poorly and insignificantly with CT fibrosis scores but strongly with PFTs. Barrier uptake and RBC transfer both correlated significantly with diffusing capacity of the lungs for carbon monoxide (r=−0.75, p<0.01 and r=0.72, p<0.01), while their ratio (RBC/barrier) correlated most strongly (r=0.94, p<0.01). RBC transfer exhibited significant anterior-posterior gravitational gradients in healthy volunteers, but not in IPF, where it was significantly impaired in the basal (p=0.02) and subpleural (p<0.01) lung. Conclusions Hyperpolarized129Xe MRI is a rapid and well-tolerated exam that provides region-specific quantification of interstitial barrier thickness and RBC transfer efficiency. With further development, it could become a robust tool for measuring disease progression and therapeutic response in patients with IPF, sensitively and non-invasively.

Hyperpolarized 129Xenon Magnetic Resonance Imaging to Quantify Regional Ventilation Differences in Mild to Moderate Asthma: A Prospective Comparison Between Semiautomated Ventilation Defect Percentage Calculation and Pulmonary Function Tests.

Objectives The aim of this study was to investigate ventilation in mild to moderate asthmatic patients and age-matched controls using hyperpolarized (HP) 129Xenon magnetic resonance imaging (MRI) and correlate findings with pulmonary function tests (PFTs). Materials and Methods This single-center, Health Insurance Portability and Accountability Act–compliant prospective study was approved by our institutional review board. Thirty subjects (10 young asthmatic patients, 26 ± 6 years; 3 males, 7 females; 10 older asthmatic patients, 64 ± 6 years; 3 males, 7 females; 10 healthy controls) were enrolled. After repeated PFTs 1 week apart, the subjects underwent 2 MRI scans within 10 minutes, inhaling 1-L volumes containing 0.5 to 1 L of 129Xe. 129Xe ventilation signal was quantified by linear binning, from which the ventilation defect percentage (VDP) was derived. Differences in VDP among subgroups and variability with age were evaluated using 1-tailed t tests. Correlation of VDP with PFTs was tested using Pearson correlation coefficient. Reproducibility of VDP was assessed using Bland-Altman plots, linear regression (R2), intraclass correlation coefficient, and concordance correlation coefficient. Results Ventilation defect percentage was significantly higher in young asthmatic patients versus young healthy subjects (8.4% ± 3.2% vs 5.6% ± 1.7%, P = 0.031), but not in older asthmatic patients versus age-matched controls (16.8% ± 10.3% vs 11.6% ± 6.6%, P = 0.13). Ventilation defect percentage was found to increase significantly with age (healthy, P = 0.05; asthmatic patients, P = 0.033). Ventilation defect percentage was highly reproducible (R2 = 0.976; intraclass correlation coefficient, 0.977; concordance correlation coefficient, 0.976) and significantly correlated with FEV1% (r = −0.42, P = 0.025), FEF25%–75% (r = −0.45, P = 0.019), FEV1/FVC (r = −0.71, P < 0.0001), FeNO (r = 0.69, P < 0.0001), and RV/TLC (r = 0.51, P = 0.0067). Bland-Altman analysis showed a bias for VDP of −0.88 ± 1.52 (FEV1%, −0.33 ± 7.18). Conclusions 129Xenon MRI is able to depict airway obstructions in mild to moderate asthma and significantly correlates with PFTs.

Lung Nodule Detection by Microdose CT Versus Chest Radiography (Standard and Dual-Energy Subtracted)

OBJECTIVE The purpose of this study was to investigate the feasibility of microdose CT using a comparable dose as for conventional chest radiographs in two planes including dual-energy subtraction for lung nodule assessment. MATERIALS AND METHODS We investigated 65 chest phantoms with 141 lung nodules, using an anthropomorphic chest phantom with artificial lung nodules. Microdose CT parameters were 80 kV and 6 mAs, with pitch of 2.2. Iterative reconstruction algorithms and an integrated circuit detector system (Stellar, Siemens Healthcare) were applied for maximum dose reduction. Maximum intensity projections (MIPs) were reconstructed. Chest radiographs were acquired in two projections with bone suppression. Four blinded radiologists interpreted the images in random order. RESULTS A soft-tissue CT kernel (I30f) delivered better sensitivities in a pilot study than a hard kernel (I70f), with respective mean (SD) sensitivities of 91.1%±2.2% versus 85.6%±5.6% (p=0.041). Nodule size was measured accurately for all kernels. Mean clustered nodule sensitivity with chest radiography was 45.7%±8.1% (with bone suppression, 46.1%±8%; p=0.94); for microdose CT, nodule sensitivity was 83.6%±9% without MIP (with additional MIP, 92.5%±6%; p<10(-3)). Individual sensitivities of microdose CT for readers 1, 2, 3, and 4 were 84.3%, 90.7%, 68.6%, and 45.0%, respectively. Sensitivities with chest radiography for readers 1, 2, 3, and 4 were 42.9%, 58.6%, 36.4%, and 90.7%, respectively. In the per-phantom analysis, respective sensitivities of microdose CT versus chest radiography were 96.2% and 75% (p<10(-6)). The effective dose for chest radiography including dual-energy subtraction was 0.242 mSv; for microdose CT, the applied dose was 0.1323 mSv. CONCLUSION Microdose CT is better than the combination of chest radiography and dual-energy subtraction for the detection of solid nodules between 5 and 12 mm at a lower dose level of 0.13 mSv. Soft-tissue kernels allow better sensitivities. These preliminary results indicate that microdose CT has the potential to replace conventional chest radiography for lung nodule detection.