BackgroundChest radiograph interpretation is critical for the detection of thoracic diseases, including tuberculosis and lung cancer, which affect millions of people worldwide each year. This time-consuming task typically requires expert radiologists to read the images, leading to fatigue-based diagnostic error and lack of diagnostic expertise in areas of the world where radiologists are not available. Recently, deep learning approaches have been able to achieve expert-level performance in medical image interpretation tasks, powered by large network architectures and fueled by the emergence of large labeled datasets. The purpose of this study is to investigate the performance of a deep learning algorithm on the detection of pathologies in chest radiographs compared with practicing radiologists.Methods and findingsWe developed CheXNeXt, a convolutional neural network to concurrently detect the presence of 14 different pathologies, including pneumonia, pleural effusion, pulmonary masses, and nodules in frontal-view chest radiographs. CheXNeXt was trained and internally validated on the ChestX-ray8 dataset, with a held-out validation set consisting of 420 images, sampled to contain at least 50 cases of each of the original pathology labels. On this validation set, the majority vote of a panel of 3 board-certified cardiothoracic specialist radiologists served as reference standard. We compared CheXNeXt’s discriminative performance on the validation set to the performance of 9 radiologists using the area under the receiver operating characteristic curve (AUC). The radiologists included 6 board-certified radiologists (average experience 12 years, range 4–28 years) and 3 senior radiology residents, from 3 academic institutions. We found that CheXNeXt achieved radiologist-level performance on 11 pathologies and did not achieve radiologist-level performance on 3 pathologies. The radiologists achieved statistically significantly higher AUC performance on cardiomegaly, emphysema, and hiatal hernia, with AUCs of 0.888 (95% confidence interval [CI] 0.863–0.910), 0.911 (95% CI 0.866–0.947), and 0.985 (95% CI 0.974–0.991), respectively, whereas CheXNeXt’s AUCs were 0.831 (95% CI 0.790–0.870), 0.704 (95% CI 0.567–0.833), and 0.851 (95% CI 0.785–0.909), respectively. CheXNeXt performed better than radiologists in detecting atelectasis, with an AUC of 0.862 (95% CI 0.825–0.895), statistically significantly higher than radiologists' AUC of 0.808 (95% CI 0.777–0.838); there were no statistically significant differences in AUCs for the other 10 pathologies. The average time to interpret the 420 images in the validation set was substantially longer for the radiologists (240 minutes) than for CheXNeXt (1.5 minutes). The main limitations of our study are that neither CheXNeXt nor the radiologists were permitted to use patient history or review prior examinations and that evaluation was limited to a dataset from a single institution.ConclusionsIn this study, we developed and validated a deep learning algorithm that classified clinically ...

BackgroundMagnetic resonance imaging (MRI) of the knee is the preferred method for diagnosing knee injuries. However, interpretation of knee MRI is time-intensive and subject to diagnostic error and variability. An automated system for interpreting knee MRI could prioritize high-risk patients and assist clinicians in making diagnoses. Deep learning methods, in being able to automatically learn layers of features, are well suited for modeling the complex relationships between medical images and their interpretations. In this study we developed a deep learning model for detecting general abnormalities and specific diagnoses (anterior cruciate ligament [ACL] tears and meniscal tears) on knee MRI exams. We then measured the effect of providing the model’s predictions to clinical experts during interpretation.Methods and findingsOur dataset consisted of 1,370 knee MRI exams performed at Stanford University Medical Center between January 1, 2001, and December 31, 2012 (mean age 38.0 years; 569 [41.5%] female patients). The majority vote of 3 musculoskeletal radiologists established reference standard labels on an internal validation set of 120 exams. We developed MRNet, a convolutional neural network for classifying MRI series and combined predictions from 3 series per exam using logistic regression. In detecting abnormalities, ACL tears, and meniscal tears, this model achieved area under the receiver operating characteristic curve (AUC) values of 0.937 (95% CI 0.895, 0.980), 0.965 (95% CI 0.938, 0.993), and 0.847 (95% CI 0.780, 0.914), respectively, on the internal validation set. We also obtained a public dataset of 917 exams with sagittal T1-weighted series and labels for ACL injury from Clinical Hospital Centre Rijeka, Croatia. On the external validation set of 183 exams, the MRNet trained on Stanford sagittal T2-weighted series achieved an AUC of 0.824 (95% CI 0.757, 0.892) in the detection of ACL injuries with no additional training, while an MRNet trained on the rest of the external data achieved an AUC of 0.911 (95% CI 0.864, 0.958). We additionally measured the specificity, sensitivity, and accuracy of 9 clinical experts (7 board-certified general radiologists and 2 orthopedic surgeons) on the internal validation set both with and without model assistance. Using a 2-sided Pearson’s chi-squared test with adjustment for multiple comparisons, we found no significant differences between the performance of the model and that of unassisted general radiologists in detecting abnormalities. General radiologists achieved significantly higher sensitivity in detecting ACL tears (p-value = 0.002; q-value = 0.019) and significantly higher specificity in detecting meniscal tears (p-value = 0.003; q-value = 0.019). Using a 1-tailed t test on the change in performance metrics, we found that providing model predictions significantly increased clinical experts’ specificity in identifying ACL tears (p-value < 0.001; q-value = 0.006). The primary limitations of our study include lack of surgical ground truth and the small size of ...

One of the earliest events in programmed cell death is the externalization of phosphatidylserine, a membrane phospholipid normally restricted to the inner leaf let of the lipid bilayer. Annexin V, an endogenous human protein with a high affinity for membrane bound phosphatidylserine, can be used in vitro to detect apoptosis before other well described morphologic or nuclear changes associated with programmed cell death. We tested the ability of exogenously administered radiolabeled annexin V to concentrate at sites of apoptotic cell death in vivo. After derivatization with hydrazinonicotinamide, annexin V was radiolabeled with technetium 99m. In vivo localization of technetium 99m hydrazinonicotinamide-annexin V was tested in three models: fuminant hepatic apoptosis induced by anti-Fas antibody injection in BALB͞c mice; acute rejection in ACI rats with transplanted heterotopic PVG cardiac allografts; and cyclophosphamide treatment of transplanted 38C13 murine B cell lymphomas. External radionuclide imaging showed a two-to sixfold increase in the uptake of radiolabeled annexin V at sites of apoptosis in all three models. Immunohistochemical staining of cardiac allografts for exogenously administered annexin V revealed intense staining of numerous myocytes at the periphery of mononuclear infiltrates of which only a few demonstrated positive apoptotic nuclei by the terminal deoxynucleotidyltransferase-mediated UTP end labeling method. These results suggest that radiolabeled annexin V can be used in vivo as a noninvasive means to detect and serially image tissues and organs undergoing programmed cell death.Programmed cell death (apoptosis) plays a crucial role in the pathogenesis of a number of disorders including AIDS and other viral illnesses, cerebral and myocardial ischemia, autoimmune and neurodegenerative diseases, organ and bone marrow transplant rejection, and tumor response to chemotherapy and radiation (1-3). Since the original description of apoptosis by Wyllie in 1972, its assessment in vivo has required direct examination of biopsied or aspirated material (4). An imaging technique capable of localizing and quantifying apoptosis in vivo would permit assessment of disease progression or regression and similarly define the efficacy of therapy designed to inhibit or induce cell death (5-6).Cells undergoing apoptosis redistribute phosphatidylserine (PS) from the inner leaflet of the plasma membrane lipid bilayer to the outer leaflet (7,8). The externalization of PS is a general feature of apoptosis occurring before membrane bleb formation and DNA degradation (7,8). Annexin V, a human protein with a molecular weight of 36,000 has a high affinity for cell or platelet membranes with exposed PS in vitro and in vivo (9-13). This observation has led to testing radiolabeled annexin V in animal models of acute thrombosis and imaging of atrial thrombi in patients with atrial fibrillation (14, 15). In the current study, annexin V was derivatized with hydrazinonicotinamide (HYNIC) and coupled to technetium 99m ...

IntroductionCD4 ϩ CD25 ϩ regulatory T (Treg) cells are potent modulators of immune responses. We and others have demonstrated that donorderived CD4 ϩ CD25 ϩ Treg cells could suppress lethal acute graft-versus-host disease (aGVHD) in murine models of allogeneic bone marrow transplantation. [1][2][3] In these models, cotransplantation of Treg cells with conventional donor T cells controlled the expansion of alloaggressive T cells in recipient animals, thereby interfering with one of the major events in the initiation phase of aGVHD. 4 Importantly, donor Treg cells did not cause generalized immune paralysis, since the beneficial graft-versus-leukemia/ lymphoma (GVL) effect of donor T cells was maintained. [4][5][6] Modulating alloimmune responses after bone marrow or hematopoietic stem cell transplantation with adoptively transferred donor CD4 ϩ CD25 ϩ Treg cells thus appears as a promising strategy for the prevention or therapy of aGVHD in humans.CD62L (L-selectin) is an important T-cell homing receptor as well as a marker for T-cell development. Naive T cells are CD62L ϩ and interaction of CD62L with its ligands, a group of molecules collectively referred to as peripheral node addressin (PNAd), is crucial for T-cell entry into lymph nodes (LNs) via high endothelial venules. 7 Expression of CD62L is rapidly lost following T-cell receptor engagement, and CD62L Ϫ T cells are thought to be "antigen experienced." There are 2 recent publications demonstrating that CD62L Ϫ donor T cells did not cause GVHD. 8,9 While CD4 ϩ CD62L Ϫ T cells contained a higher fraction of CD4 ϩ CD25 ϩ Treg cells, their inability to induce GVHD was maintained after Treg cell depletion. 8 Nevertheless, these findings called for the reciprocal analysis of the GVHD-regulating capacity of CD62L ϩ and CD62L Ϫ subsets of CD4 ϩ CD25 ϩ Treg cells.Comparing the effects CD4 ϩ CD25 ϩ CD62L ϩ and CD62L Ϫ Treg cell subpopulations in aGVHD was suggested by a second line of investigations. Both subsets had been shown to be equally anergic and suppressive upon polyclonal stimulation in vitro. [10][11][12] Interestingly, we found that in an adoptive transfer model of diabetes into nonobese diabetic-severe combined immunodeficient (NOD-scid) mice, only the CD62L ϩ but not the CD62L Ϫ subset of CD4 ϩ CD25 ϩ Treg cells caused a significant delay of disease onset. 12 In contrast, another group reported recently that both CD62L ϩ and CD62L Ϫ Treg cell subsets were protective in an adoptive transfer model of colitis, 13 suggesting that the inconsistency between in vitro and in vivo experiments in NOD mice may have been related to peculiarities of this mouse strain, which spontaneously develops autoimmune diabetes. In the current study, we used the aGVHD model as an additional in vivo assay for suppressor function in a nonautoimmune disease-prone strain.Using a mouse model for lethal aGVHD induced by major histocompatibility complex (MHC)-mismatched CD4 ϩ CD25 Ϫ T cells, 1 we found that only the CD62L ϩ subpopulation of CD4 ϩ CD25 ϩ T cells protected recipients aga...