Transfer Learning for Automated COVID-19 B-Line Classification in Lung Ultrasound.

Lung ultrasound (LUS) as a diagnostic tool is gaining support for its role in the diagnosis and management of COVID-19 and a number of other lung pathologies. B-lines are a predominant feature in COVID-19, however LUS requires a skilled clinician to interpret findings. To facilitate the interpretation, our main objective was to develop automated methods to classify B-lines as pathologic vs. normal. We developed transfer learning models based on ResNet networks to classify B-lines as pathologic (at least 3 B-lines per lung field) vs. normal using COVID-19 LUS data. Assessment of B-line severity on a 0-4 multi-class scale was also explored. For binary B-line classification, at the frame-level, all ResNet models pretrained with ImageNet yielded higher performance than the baseline nonpretrained ResNet-18. Pretrained ResNet-18 has the best Equal Error Rate (EER) of 9.1% vs the baseline of 11.9%. At the clip-level, all pretrained network models resulted in better Cohen's kappa agreement (linear-weighted) and clip score accuracy, with the pretrained ResNet-18 having the best Cohen's kappa of 0.815 [95% CI: 0.804-0.826], and ResNet-101 the best clip scoring accuracy of 93.6%. Similar results were shown for multi-class scoring, where pretrained network models outperformed the baseline model. A class activation map is also presented to guide clinicians in interpreting LUS findings. Future work aims to further improve the multi-class assessment for severity of B-lines with a more diverse LUS dataset.

Subchondral Bone Length in Knee Osteoarthritis: A Deep Learning-Derived Imaging Measure and Its Association With Radiographic and Clinical Outcomes.

OBJECTIVE: To develop a bone shape measure that reflects the extent of cartilage loss and bone flattening in knee osteoarthritis (OA) and test it against estimates of disease severity. METHODS: A fast region-based convolutional neural network was trained to crop the knee joints in sagittal dual-echo steady-state magnetic resonance imaging sequences obtained from the Osteoarthritis Initiative (OAI). Publicly available annotations of the cartilage and menisci were used as references to annotate the tibia and the femur in 61 knees. Another deep neural network (U-Net) was developed to learn these annotations. Model predictions were compared to radiologist-driven annotations on an independent test set (27 knees). The U-Net was applied to automatically extract the knee joint structures on the larger OAI data set (n = 9,434 knees). We defined subchondral bone length (SBL), a novel shape measure characterizing the extent of overlying cartilage and bone flattening, and examined its relationship with radiographic joint space narrowing (JSN), concurrent pain and disability (according to the Western Ontario and McMaster Universities Osteoarthritis Index), as well as subsequent partial or total knee replacement. Odds ratios (ORs) and 95% confidence intervals (95% CIs) for each outcome were estimated using relative changes in SBL from the OAI data set stratified into quartiles. RESULTS: The mean SBL values for knees with JSN were consistently different from knees without JSN. Greater changes of SBL from baseline were associated with greater pain and disability. For knees with medial or lateral JSN, the ORs for future knee replacement between the lowest and highest quartiles corresponding to SBL changes were 5.68 (95% CI 3.90-8.27) and 7.19 (95% CI 3.71-13.95), respectively. CONCLUSION: SBL quantified OA status based on JSN severity and shows promise as an imaging marker in predicting clinical and structural OA outcomes.

Machine learning combining CT findings and clinical parameters improves prediction of length of stay and ICU admission in torso trauma.

OBJECTIVE: To develop machine learning (ML) models capable of predicting ICU admission and extended length of stay (LOS) after torso (chest, abdomen, or pelvis) trauma, by using clinical and/or imaging data. MATERIALS AND METHODS: This was a retrospective study of 840 adult patients admitted to a level 1 trauma center after injury to the torso over the course of 1 year. Clinical parameters included age, sex, vital signs, clinical scores, and laboratory values. Imaging data consisted of any injury present on CT. The two outcomes of interest were ICU admission and extended LOS, defined as more than the median LOS in the dataset. We developed and tested artificial neural network (ANN) and support vector machine (SVM) models, and predictive performance was evaluated by area under the receiver operating characteristic (ROC) curve (AUC). RESULTS: The AUCs of SVM and ANN models to predict ICU admission were up to 0.87 ± 0.03 and 0.78 ± 0.02, respectively. The AUCs of SVM and ANN models to predict extended LOS were up to 0.80 ± 0.04 and 0.81 ± 0.05, respectively. Predictions based on imaging alone or imaging with clinical parameters were consistently more accurate than those based solely on clinical parameters. CONCLUSIONS: The best performing models incorporated imaging findings and outperformed those with clinical findings alone. ML models have the potential to help predict outcomes in trauma by integrating clinical and imaging findings, although further research may be needed to optimize their performance. KEY POINTS: • Artificial neural network and support vector machine-based models were used to predict the intensive care unit admission and extended length of stay after trauma to the torso. • Our input data consisted of clinical parameters and CT imaging findings derived from radiology reports, and we found that combining the two significantly enhanced the prediction of both outcomes with either model. • The highest accuracy (83%) and highest area under the receiver operating characteristic curve (0.87) were obtained for artificial neural networks and support vector machines, respectively, by combining clinical and imaging features in the prediction of intensive care unit admission.

Detecting Intracranial Hemorrhage with Deep Learning.

Initial results are reported on automated detection of intracranial hemorrhage from CT, which would be valuable in a computer-aided diagnosis system to help the radiologist detect subtle hemorrhages. Previous work has taken a classic approach involving multiple steps of alignment, image processing, image corrections, handcrafted feature extraction, and classification. Our current work instead uses a deep convolutional neural network to simultaneously learn features and classification, eliminating the multiple hand-tuned steps. Performance is improved by computing the mean output for rotations of the input image. Postprocessing is additionally applied to the CNN output to significantly improve specificity. The database consists of 134 CT cases (4,300 images), divided into 60, 5, and 69 cases for training, validation, and test. Each case typically includes multiple hemorrhages. Performance on the test set was 81% sensitivity per lesion (34/42 lesions) and 98% specificity per case (45/46 cases). The sensitivity is comparable to previous results (on different datasets), but with a significantly higher specificity. In addition, insights are shared to improve performance as the database is expanded.

Breath of relief: apparent pulmonary nodule due to heterotopic ossification.

Extrapulmonary heterotopic ossification appears similarly to pulmonary nodules on CXR, and is in the differential for pulmonary nodules. It occurs following the bone trauma, and in early stages appears similarly to tumors. Heterotopic ossification is diagnosed by its calcification pattern via MRI or ultrasound and managed conservatively unless symptoms develop.

Texture analysis as a radiomic marker for differentiating renal tumors.

PURPOSE: To evaluate the utility of texture analysis for the differentiation of renal tumors, including the various renal cell carcinoma subtypes and oncocytoma. MATERIALS AND METHODS: Following IRB approval, a retrospective analysis was performed, including all patients with pathology-proven renal tumors and an abdominal computed tomography (CT) examination. CT images of the tumors were manually segmented, and texture analysis of the segmented tumors was performed. A support vector machine (SVM) method was also applied to classify tumor types. Texture analysis results were compared to the various tumors and areas under the curve (AUC) were calculated. Similar calculations were performed with the SVM data. RESULTS: One hundred nineteen patients were included. Excellent discriminators of tumors were identified among the histogram-based features noting features skewness and kurtosis, which demonstrated AUCs of 0.91 and 0.93 (p < 0.0001), respectively, for differentiating clear cell subtype from oncocytoma. Histogram feature median demonstrated an AUC of 0.99 (p < 0.0001) for differentiating papillary subtype from oncocytoma and an AUC of 0.92 for differentiating oncocytoma from other tumors. Machine learning further improved the results achieving very good to excellent discrimination of tumor subtypes. The ability of machine learning to distinguish clear cell subtype from other tumors and papillary subtype from other tumors was excellent with AUCs of 0.91 and 0.92, respectively. CONCLUSION: Texture analysis is a promising non-invasive tool for distinguishing renal tumors on CT images. These results were further improved upon application of machine learning, and support the further development of texture analysis as a quantitative biomarker for distinguishing various renal tumors.

Characterization of gradient echo signal decays in healthy and cancerous prostate at 3T improves with a Gaussian augmentation of the mono-exponential (GAME) model.

A biomarker of cancer aggressiveness, such as hypoxia, could substantially impact treatment decisions in the prostate, especially radiation therapy, by balancing treatment morbidity (urinary incontinence, erectile dysfunction, etc.) against mortality. R2 (*) mapping with Mono-Exponential (ME) decay modeling has shown potential for identifying areas of prostate cancer hypoxia at 1.5T. However, Gaussian deviations from ME decay have been observed in other tissues at 3T. The purpose of this study is to assess whether gradient-echo signal decays are better characterized by a standard ME decay model, or a Gaussian Augmentation of the Mono-Exponential (GAME) decay model, in the prostate at 3T. Multi-gradient-echo signals were acquired on 20 consecutive patients with a clinical suspicion of prostate cancer undergoing MR-guided prostate biopsies. Data were fitted with both ME and GAME models. The information contents of these models were compared using Akaike's information criterion (second order, AICC ), in skeletal muscle, the prostate central gland (CG), and peripheral zone (PZ) regions of interest (ROIs). The GAME model had higher information content in 30% of the prostate on average (across all patients and ROIs), covering up to 67% of cancerous PZ ROIs, and up to 100% of cancerous CG ROIs (in individual patients). The higher information content of GAME became more prominent in regions that would be assumed hypoxic using ME alone, reaching 50% of the PZ and 70% of the CG as ME R2 (*) approached 40 s(-1) . R2 (*) mapping may have important applications in MRI; however, information lost due to modeling could mask differences in parameters due to underlying tissue anatomy or physiology. The GAME model improves characterization of signal behavior in the prostate at 3T, and may increase the potential for determining correlates of fit parameters with biomarkers, for example of oxygenation status.

Multiexponential T(2) analyses in a murine model of hepatic fibrosis at 11.7 T MRI.

This study evaluated the effects of hepatic fibrosis on the multiexponential T(2) (MET(2) ) relaxation of ex vivo murine liver specimens using an 11.7 T MRI. This animal study was approved by the Institutional Animal Care and Use Committee. Eighteen male C57BL/6 mice were divided into control (n = 3) and experimental (n = 15) groups; the latter group was fed a 3,5-dicarbethoxy-1,4-dihydrocollidine-supplemented diet to induce hepatic fibrosis. Ex vivo liver specimens were imaged using an 11.7 T MRI scanner. A multi-echo spin-echo sequence was utilized for subsequent MET(2) analysis. Degrees of fibrosis were determined by a pathologist, as well as by digital image analysis. Scatterplot graphs comparing various features of the MET(2) signal decay with the degrees of fibrosis were generated, and correlation coefficients were calculated. Two distinct peaks of the MET(2) signal decay were identified in all liver specimens: a short T(2) component with a geometric mean T(2) (GMT(2) ) approximating 30 ms; and a long T(2) component with GMT(2) approximating 400 ms. Strong correlation was found between the degree of hepatic fibrosis and the amplitude of the short T(2) component, with a higher degrees of fibrosis associated with a lower amplitude. Moderate correlation was also found between hepatic fibrosis and the GMT(2) values of the long T(2) component, with higher degrees of fibrosis associated with lower GMT(2) values. The study of hepatic microenvironments using MET(2) analyses offers potential utility in the ongoing development of the noninvasive assessment of hepatic fibrosis using MRI.

Quantitative MR imaging: physical principles and sequence design in abdominal imaging.

Quantitative magnetic resonance (MR) imaging seeks to quantify fundamental biologic and MR-inducible tissue properties, in contrast to the routine application of MR imaging in the clinic, in which differences in MR parameters are used to generate contrast for subsequent subjective image analysis. Fundamental parameters that are commonly quantified by using MR imaging include proton density, diffusion, T1 relaxation, T2 and T2* relaxation, and magnetization transfer. Applications of these MR imaging-quantifiable parameters to abdominal imaging include oncologic imaging, evaluation of diffuse liver disease, and assessment of splenic, renal, and pancreatic disease. An understanding of the inherent physical principles underlying the basic quantitative parameters as well as the commonly used pulse sequences requisite to their derivation is critical, as this field is rapidly growing and its use will likely continue to expand in the clinic. The full potential of quantitative MR imaging applied to abdominal imaging has yet to be realized, but the myriad applications reported to date will undoubtedly continue to grow.