Deep learning workflow in radiology: a primer Montagnon, Emmanuel; Cerny, Milena; Cadrin-Chênevert, Alexandre ...
Insights into imaging,
02/2020, Letnik:
11, Številka:
1
Journal Article
Recenzirano
Odprti dostop
Interest for deep learning in radiology has increased tremendously in the past decade due to the high achievable performance for various computer vision tasks such as detection, segmentation, ...classification, monitoring, and prediction. This article provides step-by-step practical guidance for conducting a project that involves deep learning in radiology, from defining specifications, to deployment and scaling. Specifically, the objectives of this article are to provide an overview of clinical use cases of deep learning, describe the composition of multi-disciplinary team, and summarize current approaches to patient, data, model, and hardware selection. Key ideas will be illustrated by examples from a prototypical project on imaging of colorectal liver metastasis. This article illustrates the workflow for liver lesion detection, segmentation, classification, monitoring, and prediction of tumor recurrence and patient survival. Challenges are discussed, including ethical considerations, cohorting, data collection, anonymization, and availability of expert annotations. The practical guidance may be adapted to any project that requires automated medical image analysis.
Abstract We aimed to implement four data partitioning strategies evaluated with four federated learning (FL) algorithms and investigate the impact of data distribution on FL model performance in ...detecting steatosis using B-mode US images . A private dataset (153 patients; 1530 images) and a public dataset (55 patient; 550 images) were included in this retrospective study. The datasets contained patients with metabolic dysfunction-associated fatty liver disease (MAFLD) with biopsy-proven steatosis grades and control individuals without steatosis. We employed four data partitioning strategies to simulate FL scenarios and we assessed four FL algorithms. We investigated the impact of class imbalance and the mismatch between the global and local data distributions on the learning outcome. Classification performance was assessed with area under the receiver operating characteristic curve (AUC) on a separate test set. AUCs were 0.93 (95% CI 0.92, 0.94) for source-based partitioning scenario with FedAvg, 0.90 (95% CI 0.89, 0.91) for a centralized model, and 0.83 (95% CI 0.81, 0.85) for a model trained in a single-center scenario. When data was perfectly balanced on the global level and each site had an identical data distribution, the model yielded an AUC of 0.90 (95% CI 0.88, 0.92). When each site contained data exclusively from one single class, irrespective of the global data distribution, the AUC fell in the range of 0.34–0.70. FL applied to B-mode US images provide performance comparable to a centralized model and higher than single-center scenario. Global data imbalance and local data heterogeneity influenced the learning outcome.
Finding a noninvasive radiomic surrogate of tumor immune features could help identify patients more likely to respond to novel immune checkpoint inhibitors. Particularly, CD73 is an ectonucleotidase ...that catalyzes the breakdown of extracellular AMP into immunosuppressive adenosine, which can be blocked by therapeutic antibodies. High CD73 expression in colorectal cancer liver metastasis (CRLM) resected with curative intent is associated with early recurrence and shorter patient survival. The aim of this study was hence to evaluate whether machine learning analysis of preoperative liver CT-scan could estimate high vs low CD73 expression in CRLM and whether such radiomic score would have a prognostic significance.
We trained an Attentive Interpretable Tabular Learning (TabNet) model to predict, from preoperative CT images, stratified expression levels of CD73 (CD73
vs. CD73
) assessed by immunofluorescence (IF) on tissue microarrays. Radiomic features were extracted from 160 segmented CRLM of 122 patients with matched IF data, preprocessed and used to train the predictive model. We applied a five-fold cross-validation and validated the performance on a hold-out test set.
TabNet provided areas under the receiver operating characteristic curve of 0.95 (95% CI 0.87 to 1.0) and 0.79 (0.65 to 0.92) on the training and hold-out test sets respectively, and outperformed other machine learning models. The TabNet-derived score, termed rad-CD73, was positively correlated with CD73 histological expression in matched CRLM (Spearman's ρ = 0.6004; P < 0.0001). The median time to recurrence (TTR) and disease-specific survival (DSS) after CRLM resection in rad-CD73
vs rad-CD73
patients was 13.0 vs 23.6 months (P = 0.0098) and 53.4 vs 126.0 months (P = 0.0222), respectively. The prognostic value of rad-CD73 was independent of the standard clinical risk score, for both TTR (HR = 2.11, 95% CI 1.30 to 3.45, P < 0.005) and DSS (HR = 1.88, 95% CI 1.11 to 3.18, P = 0.020).
Our findings reveal promising results for non-invasive CT-scan-based prediction of CD73 expression in CRLM and warrant further validation as to whether rad-CD73 could assist oncologists as a biomarker of prognosis and response to immunotherapies targeting the adenosine pathway.
In developed countries, colorectal cancer is the second cause of cancer-related mortality. Chemotherapy is considered a standard treatment for colorectal liver metastases (CLM). Among patients who ...develop CLM, the assessment of patient response to chemotherapy is often required to determine the need for second-line chemotherapy and eligibility for surgery. However, while FOLFOX-based regimens are typically used for CLM treatment, the identification of responsive patients remains elusive. Computer-aided diagnosis systems may provide insight in the classification of liver metastases identified on diagnostic images. In this paper, we propose a fully automated framework based on deep convolutional neural networks (DCNN) which first differentiates treated and untreated lesions to identify new lesions appearing on CT scans, followed by a fully connected neural networks to predict from untreated lesions in pre-treatment computed tomography (CT) for patients with CLM undergoing chemotherapy, their response to a FOLFOX with Bevacizumab regimen as first-line of treatment. The ground truth for assessment of treatment response was histopathology-determined tumor regression grade. Our DCNN approach trained on 444 lesions from 202 patients achieved accuracies of 91% for differentiating treated and untreated lesions, and 78% for predicting the response to FOLFOX-based chemotherapy regimen. Experimental results showed that our method outperformed traditional machine learning algorithms and may allow for the early detection of non-responsive patients.
Multiple clinical phenotypes have been proposed for coronavirus disease (COVID-19), but few have used multimodal data. Using clinical and imaging data, we aimed to identify distinct clinical ...phenotypes in patients admitted with COVID-19 and to assess their clinical outcomes. Our secondary objective was to demonstrate the clinical applicability of this method by developing an interpretable model for phenotype assignment.
We analyzed data from 547 patients hospitalized with COVID-19 at a Canadian academic hospital. We processed the data by applying a factor analysis of mixed data (FAMD) and compared four clustering algorithms: k-means, partitioning around medoids (PAM), and divisive and agglomerative hierarchical clustering. We used imaging data and 34 clinical variables collected within the first 24 h of admission to train our algorithm. We conducted a survival analysis to compare the clinical outcomes across phenotypes. With the data split into training and validation sets (75/25 ratio), we developed a decision-tree-based model to facilitate the interpretation and assignment of the observed phenotypes.
Agglomerative hierarchical clustering was the most robust algorithm. We identified three clinical phenotypes: 79 patients (14%) in Cluster 1, 275 patients (50%) in Cluster 2, and 203 (37%) in Cluster 3. Cluster 2 and Cluster 3 were both characterized by a low-risk respiratory and inflammatory profile but differed in terms of demographics. Compared with Cluster 3, Cluster 2 comprised older patients with more comorbidities. Cluster 1 represented the group with the most severe clinical presentation, as inferred by the highest rate of hypoxemia and the highest radiological burden. Intensive care unit (ICU) admission and mechanical ventilation risks were the highest in Cluster 1. Using only two to four decision rules, the classification and regression tree (CART) phenotype assignment model achieved an AUC of 84% (81.5-86.5%, 95 CI) on the validation set.
We conducted a multidimensional phenotypic analysis of adult inpatients with COVID-19 and identified three distinct phenotypes associated with different clinical outcomes. We also demonstrated the clinical usability of this approach, as phenotypes can be accurately assigned using a simple decision tree. Further research is still needed to properly incorporate these phenotypes in the management of patients with COVID-19.
Stolt's f-k migration for plane wave ultrasound imaging Garcia, D.; Tarnec, L. L.; Muth, S. ...
IEEE transactions on ultrasonics, ferroelectrics and frequency control/IEEE transactions on ultrasonics, ferroelectrics, and frequency control,
09/2013, Letnik:
60, Številka:
9
Journal Article
Recenzirano
Odprti dostop
Ultrafast ultrasound is an emerging modality that offers new perspectives and opportunities in medical imaging. Plane wave imaging (PWI) allows one to attain very high frame rates by transmission of ...planar ultrasound wavefronts. As a plane wave reaches a given scatterer, the latter becomes a secondary source emitting upward spherical waves and creating a diffraction hyperbola in the received RF signals. To produce an image of the scatterers, all the hyperbolas must be migrated back to their apexes. To perform beamforming of plane wave echo RFs and return high-quality images at high frame rates, we propose a new migration method carried out in the frequency-wavenumber (f-k) domain. The f-k migration for PWI has been adapted from the Stolt migration for seismic imaging. This migration technique is based on the exploding reflector model (ERM), which consists in assuming that all the scatterers explode in concert and become acoustic sources. The classical ERM model, however, is not appropriate for PWI. We showed that the ERM can be made suitable for PWI by a spatial transformation of the hyperbolic traces present in the RF data. In vitro experiments were performed to outline the advantages of PWI with Stolt's f-k migration over the conventional delay-and-sum (DAS) approach. The Stolt's f-k migration was also compared with the Fourier-based method developed by J.-Y. Lu. Our findings show that multi-angle compounded f-k migrated images are of quality similar to those obtained with a stateof- the-art dynamic focusing mode. This remained true even with a very small number of steering angles, thus ensuring a highly competitive frame rate. In addition, the new FFT-based f-k migration provides comparable or better contrast-to-noise ratio and lateral resolution than the Lu's and DAS migration schemes. Matlab codes for the Stolt's f-k migration for PWI are provided.
A method based on adaptive torsional shear waves (ATSW) is proposed to overcome the strong attenuation of shear waves generated by a radiation force in dynamic elastography. During the inward ...propagation of ATSW, the magnitude of displacements is enhanced due to the convergence of shear waves and constructive interferences. The proposed method consists in generating ATSW fields from the combination of quasi-plane shear wavefronts by considering a linear superposition of displacement maps. Adaptive torsional shear waves were experimentally generated in homogeneous and heterogeneous tissue mimicking phantoms, and compared to quasi-plane shear wave propagations. Results demonstrated that displacement magnitudes by ATSW could be up to 3 times higher than those obtained with quasi-plane shear waves, that the variability of shear wave speeds was reduced, and that the signal-to-noise ratio of displacements was improved. It was also observed that ATSW could cause mechanical inclusions to resonate in heterogeneous phantoms, which further increased the displacement contrast between the inclusion and the surrounding medium. This method opens a way for the development of new noninvasive tissue characterization strategies based on ATSW in the framework of our previously reported shear wave induced resonance elastography (SWIRE) method proposed for breast cancer diagnosis.
Deep vein thrombosis is a common vascular disease that can lead to pulmonary embolism and death. The early diagnosis and clot age staging are important parameters for reliable therapy planning. This ...article presents an acoustic radiation force induced resonance elastography method for the viscoelastic characterization of clotting blood. The physical concept of this method relies on the mechanical resonance of the blood clot occurring at specific frequencies. Resonances are induced by focusing ultrasound beams inside the sample under investigation. Coupled to an analytical model of wave scattering, the ability of the proposed method to characterize the viscoelasticity of a mimicked venous thrombosis in the acute phase is demonstrated. Experiments with a gelatin-agar inclusion sample of known viscoelasticity are performed for validation and establishment of the proof of concept. In addition, an inversion method is applied in vitro for the kinetic monitoring of the blood coagulation process of six human blood samples obtained from two volunteers. The computed elasticity and viscosity values of blood samples at the end of the 90 min kinetics were estimated at 411 ± 71 Pa and 0.25 ± 0.03 Pa · s for volunteer #1, and 387 ± 35 Pa and 0.23 ± 0.02 Pa · s for volunteer #2, respectively. The proposed method allowed reproducible time-varying thrombus viscoelastic measurements from samples having physiological dimensions.
To assess the reproducibility of six ultrasound (US)-determined shear wave (SW) viscoelastography parameters for assessment of mechanical properties of the liver in volunteers and patients with ...biopsy-proven metabolic dysfunction-associated steatotic liver disease (MASLD) or metabolic dysfunction-associated steatohepatitis (MASH).OBJECTIVETo assess the reproducibility of six ultrasound (US)-determined shear wave (SW) viscoelastography parameters for assessment of mechanical properties of the liver in volunteers and patients with biopsy-proven metabolic dysfunction-associated steatotic liver disease (MASLD) or metabolic dysfunction-associated steatohepatitis (MASH).This prospective, cross-sectional, institutional review board-approved study included 10 volunteers and 20 patients with MASLD or MASH who underwent liver US elastography twice, at least 2 weeks apart. SW speed (SWS), Young's modulus (E), shear modulus (G), SW attenuation (SWA), SW dispersion (SWD), and viscosity were computed from radiofrequency data recorded on a research US scanner. Linear mixed models were used to consider the sonographer on duty as a confounder. The reproducibility of measurements was assessed by intraclass correlation coefficient (ICC), coefficient of variation (CV), reproducibility coefficient (RDC), and Bland-Altman analyses.METHODSThis prospective, cross-sectional, institutional review board-approved study included 10 volunteers and 20 patients with MASLD or MASH who underwent liver US elastography twice, at least 2 weeks apart. SW speed (SWS), Young's modulus (E), shear modulus (G), SW attenuation (SWA), SW dispersion (SWD), and viscosity were computed from radiofrequency data recorded on a research US scanner. Linear mixed models were used to consider the sonographer on duty as a confounder. The reproducibility of measurements was assessed by intraclass correlation coefficient (ICC), coefficient of variation (CV), reproducibility coefficient (RDC), and Bland-Altman analyses.The sonographer performing the exam had no impact on viscoelastic parameters (P > .05). ICCs of SWS, E, G, SWA, SWD, and viscosity were, respectively, 0.89 (95% confidence intervals CI: 0.79-0.95), 0.81 (95% CI: 0.79-0.95), 0.90 (95% CI: 0.80-0.95), 0.96 (95% CI: 0.93-0.98), 0.78 (95% CI: 0.60-0.89), and 0.90 (95% CI: 0.80-0.95); CVs were 11.9, 23.3, 24.2, 10.1, 29.0, and 32.2%; RDCs were 33.0, 64.5, 66.9, 27.7, 80.3, and 89.2%, and Bland-Altman mean biases and 95% limits of agreement were -0.05 (-0.45, 0.35) m/s, -0.61 (-5.33, 4.10) kPa, -0.25 (-2.06, 1.56) kPa, -0.01 (-0.27, 0.26) Np/m/Hz, -0.09 (-7.09, 6.91) m/s/kHz, and -0.33 (-2.60, 1.94) Pa/s, between the two visits.RESULTSThe sonographer performing the exam had no impact on viscoelastic parameters (P > .05). ICCs of SWS, E, G, SWA, SWD, and viscosity were, respectively, 0.89 (95% confidence intervals CI: 0.79-0.95), 0.81 (95% CI: 0.79-0.95), 0.90 (95% CI: 0.80-0.95), 0.96 (95% CI: 0.93-0.98), 0.78 (95% CI: 0.60-0.89), and 0.90 (95% CI: 0.80-0.95); CVs were 11.9, 23.3, 24.2, 10.1, 29.0, and 32.2%; RDCs were 33.0, 64.5, 66.9, 27.7, 80.3, and 89.2%, and Bland-Altman mean biases and 95% limits of agreement were -0.05 (-0.45, 0.35) m/s, -0.61 (-5.33, 4.10) kPa, -0.25 (-2.06, 1.56) kPa, -0.01 (-0.27, 0.26) Np/m/Hz, -0.09 (-7.09, 6.91) m/s/kHz, and -0.33 (-2.60, 1.94) Pa/s, between the two visits.US-determined viscoelastography parameters can be measured with high reproducibility and consistency between two visits 2 weeks apart on the same ultrasound machine.CONCLUSIONUS-determined viscoelastography parameters can be measured with high reproducibility and consistency between two visits 2 weeks apart on the same ultrasound machine.
Background
Abdominal aortic aneurysm (AAA) rupture prediction based on sex and diameter could be improved. The goal was to assess whether aortic calcification distribution could better predict AAA ...rupture through machine learning and LASSO regression.
Methodology
In this retrospective study, 80 patients treated for a ruptured AAA between January 2001 and August 2018 were matched with 80 non-ruptured patients based on maximal AAA diameter, age, and sex. Calcification volume and dispersion, morphologic, and clinical variables were compared between both groups using a univariable analysis with
p
= 0.05 and multivariable analysis through machine learning and LASSO regression. We used AUC for machine learning and odds ratios for regression to measure performance.
Results
Mean age of patients was 74.0 ± 8.4 years and 89% were men. AAA diameters were equivalent in both groups (80.9 ± 17.5 vs 79.0 ± 17.3 mm,
p
= 0.505). Ruptured aneurysms contained a smaller number of calcification aggregates (18.0 ± 17.9 vs 25.6 ± 18.9,
p
= 0.010) and were less likely to have a proximal neck (45.0% vs 76.3%,
p
< 0.001). In the machine learning analysis, 5 variables were associated to AAA rupture: proximal neck, antiplatelet use, calcification number, Euclidian distance between calcifications, and standard deviation of the Euclidian distance. A follow-up LASSO regression was concomitant with the findings of the machine learning analysis regarding calcification dispersion but discordant on calcification number.
Conclusion
There might be more to AAA calcifications that what is known in the present literature. We need larger prospective studies to investigate if indeed, calcification dispersion affects rupture risk.
Clinical relevance statement
Ruptured aneurysms are possibly more likely to have their calcification volume concentrated in a smaller geographical area.
Key Points
• Abdominal aortic aneurysm (AAA) rupture prediction based on sex and diameter could be improved.
• For a given calcification volume, AAAs with well-distributed calcification clusters could be less likely to rupture.
• A machine learning model including AAA calcifications better predicts rupture compared to a model based solely on maximal diameter and sex alone, although it might be prone to overfitting.