ABSTRACT
Owing to the superior performances, exemplar-based methods with knowledge distillation (KD) are widely applied in class incremental learning (CIL). However, it suffers from two drawbacks: 1) data imbalance between the old/learned and new classes causes the bias of the new classifier toward the head/new classes and 2) deep neural networks (DNNs) suffer from distribution drift when learning sequence tasks, which results in narrowed feature space and deficient representation of old tasks. For the first problem, we analyze the insufficiency of softmax loss when dealing with the problem of data imbalance in theory and then propose the imbalance softmax (im-softmax) loss to relieve the imbalanced data learning, where we re-scale the output logits to underfit the head/new classes. For another problem, we calibrate the feature space by incremental-adaptive angular margin (IAAM) loss. The new classes form a complete distribution in feature space yet the old are squeezed. To recover the old feature space, we first compute the included angle of normalized features and normalized anchor prototypes, and use the angle distribution to represent the class distribution, then we replenish the old distribution with the deviation from the new. Each anchor prototype is predefined as a learnable vector for a designated class. The proposed im-softmax reduces the bias in the linear classification layer. IAAM rectifies the representation learning, reduces the intra-class distance, and enlarges the inter-class margin. Finally, we seamlessly combine the im-softmax and IAAM in an end-to-end training framework, called the dual balanced class incremental learning (DBL), for further improvements. Experiments demonstrate the proposed method achieves state-of-the-art (SOTA) performances on several benchmarks, such as CIFAR10, CIFAR100, Tiny-ImageNet, and ImageNet-100.
ABSTRACT
Background: Patients with severe posttuberculous (TB) kyphosis might suffer from late-onset neurological deficits, and surgical correction may improve neurological function. However, there is a lack of predictive factors for neurological function in these patients. Objective: This study was aimed at identifying the risk factors for late-onset neurological deficits in spinal TB patients at initial and final assessments. Methods: Seventy-eight patients with severe kyphosis caused by old thoracic tuberculosis were retrospectively analyzed. Patients with active spinal TB and other spinal diseases were excluded from the analysis. The kyphosis Cobb angle, sagittal deformity angular ratio (S-DAR), and level of apex were measured and calculated on X-ray. The spinal cord cross-sectional area ratio (CSAR), spinal cord sagittal diameter ratio (SDR), and spinal cord angle (SCA) were measured on preoperative T2-weighted magnetic resonance imaging (MRI). According to the American Spinal Injury Association (ASIA) Impairment Scale (AIS) at the time of admission, the patients were divided into the symptomatic group (N = 60 patients, AIS grades A to D) and the asymptomatic group (N = 18 patients, AIS grade E). All of the symptomatic patients underwent surgery, and the patients from both groups had at least 2 years of follow-up. Relationships among the radiological parameters and initial and final AIS grades were evaluated via univariate and multivariate analyses. Results: The mean duration of kyphotic deformity was 37.4 years in the symptomatic group. There were no significant differences between the two groups in terms of CSAR, kyphosis Cobb angle, S-DAR, level of apex, or the segments that were involved. Patients from the symptomatic group exhibited significantly greater SDR and smaller SCA than those from the asymptomatic group (p < 0.01 and p < 0.01, respectively). The multivariate logistic regression identified SDR and SCA as independent factors influencing the likelihood of spinal cord injury at the initial and final assessments. Conclusions: Severe posttuberculous kyphosis may lead to significant neurological symptoms many years following the initial treatment. The predictive factors for late-onset neurological deficits include larger SDR and smaller SCA.
Subject(s)
Kyphosis , Spinal Cord Injuries , Humans , Kyphosis/diagnostic imaging , Kyphosis/surgery , Radiography , Retrospective Studies , Spinal Cord Injuries/complications , SpineABSTRACT
OBJECTIVE: Subarachnoid hemorrhage (SAH) is a rare complication secondary to cerebrospinal fluid (CSF) leakage during spinal surgery, but the specific cause of bleeding is still unclear. In this retrospective single-center study, we studied cases of acute SAH after spinal surgery to identify the related risk factors for bleeding. METHODS: A total of 441 patients with CSF leakage who underwent spinal surgery from 2011 to 2020 were retrospectively analyzed. According to whether postoperative SAH occurred, the patients were divided into the SAH group and the control group. By comparing the demographic data, past history, imaging data, intraoperative findings, perioperative complications, and treatment conditions in the 2 groups to identify the risk factors for SAH, we aimed to provide guidance for the prevention of SAH after spinal surgery. RESULTS: In our study, the incidence of CSF during spinal surgery is 3.04%, and the rate for SAH associated with spine operation is 0.16%. In the SAH group, 17 patients (73.9%) had preoperative hypertension, and 3 patients (13.0%) had diabetes. The intraoperative CSF leakage was approximately 118.4 ± 56.9 mL. The mean postoperative drainage was 15.4 ± 5.8 mL/hour. Compared with the SAH group, the control group had 123 patients (29.4%) with hypertension before the operation and 40 patients (9.6%) with diabetes. The intraoperative CSF leakage was approximately 76.3 ± 23.0 mL, and the mean postoperative continuous drainage was 9.7 ± 2.1 mL/hour. Binary logistic regression analysis showed that hypertension, intraoperative CSF leakage, and postoperative CSF continuous drainage speed were related to SAH. CONCLUSIONS: The rapid leakage of CSF caused by rupture of the dural sac during surgery and the rapid drainage of CSF after surgery were closely related to the occurrence of SAH. In addition, hypertension was a factor related to SAH during spinal surgery.
Subject(s)
Subarachnoid Hemorrhage , Humans , Neurosurgical Procedures/adverse effects , Postoperative Complications/epidemiology , Postoperative Complications/etiology , Postoperative Complications/prevention & control , Retrospective Studies , Risk Factors , Subarachnoid Hemorrhage/diagnostic imaging , Subarachnoid Hemorrhage/epidemiology , Subarachnoid Hemorrhage/etiologyABSTRACT
Although temozolomide (TMZ) is the first-line chemotherapeutic agent for glioblastoma, it is often non-curative due to drug resistance. To overcome the resistance of glioblastoma cells to TMZ, it is imperative to identify prognostic markers for outcome prediction and to develop chemo-sensitizing agents. Here, the gene expression profiles of TMZ-resistant and TMZ-sensitive samples were compared by microarray analysis, and mitogen-activated protein kinase kinase 2 (MEK2) was upregulated specifically in resistant glioma cells but not in sensitive tumor cells or non-tumor tissues. Moreover, a comprehensive analysis of patient data revealed that the increased level of MEK2 expression correlated well with the advancement of glioma grade and worse prognosis in response to TMZ treatment. Furthermore, reducing the level of MEK2 in U251 glioma cell lines or xenografted glioma models through shRNA-mediated gene knockdown inhibited cell proliferation and enhanced the sensitivity of cells toward TMZ treatment. Further analysis of tumor samples from glioma patients by real-time PCR indicated that an increased MEK2 expression level was closely associated with the activation of many drug resistance genes. Finally, these resistance genes were downregulated after MEK2 was silenced in vitro, suggesting that the mechanism of MEK2-induced chemo-resistance could be mediated by the transcriptional activation of these resistance genes. Collectively, our data indicated that the expression level of MEK2 could serve as a prognostic marker for glioma chemotherapy and that MEK2 antagonists can be used as chemo-sensitizers to enhance the treatment efficacy of TMZ.
Subject(s)
Biomarkers, Tumor/metabolism , Dacarbazine/analogs & derivatives , Glioma/drug therapy , Glioma/enzymology , MAP Kinase Kinase 2/metabolism , Molecular Targeted Therapy , Animals , Cell Line, Tumor , Cell Proliferation/drug effects , Dacarbazine/pharmacology , Dacarbazine/therapeutic use , Drug Resistance, Neoplasm/drug effects , Drug Resistance, Neoplasm/genetics , Gene Knockdown Techniques , Humans , MAP Kinase Kinase 2/antagonists & inhibitors , Male , Mice, Inbred BALB C , Mice, Nude , Neoplasm Grading , Prognosis , Temozolomide , Treatment Outcome , Up-Regulation/drug effectsABSTRACT
BACKGROUND: Angiogenesis is an important process facilitating the growth of glioblastoma (GBM). It also has drawn great attention in the treatment of GBM. GBM angiogenesis is closely related to the function of endothelial cells. microRNAs can affect the activities of endothelial 10 cells directly, or indirectly through the interaction of tumor cells and endothelial cells. However, the mechanism underlying the interaction of GBM cells regulated by specific microRNA with endothelial cells and following angiogenesis requires further research. In published articles, microRNA-16 acted as a tumor suppressor in multiple types of cancers including glioma, but the role in glioma angiogenesis has not been well elucidated. METHODS: The expression of microRNA-16 was detected in human GBM samples and normal brain tissues. microRNA-16 was transfected to GBM cell line U87 and A172 then the function of endothelial cells co-cultured with U87/A172 (miR-16 or control) were observed in vitro. Expression of VEGF family in vitro and the effect of microRNA-16 on GBM angiogenesis in vivo were also investigated. RESULTS: microRNA-16 is down-regulated in human GBM samples in contrast to the normal brain tissues. Overexpression of microRNA- 16 in the A172 and U87 GBM cell lines inhibited the activities of co-cultured endothelial cells, including proliferation, migration, extension and tubule formation. Further experiments of dual luciferase assays verified microRNA-16 directly targeting Bmi-1. microRNA-16 down-regulated the expression of vascular endothelial growth factor VEGF-A and VEGF- C which were closely related to the angiogenesis of GBM. Moreover, less vascular formed in the section of neoplasm of the microRNA- transduced group than the control group in vivo. CONCLUSIONS: Collectively, these findings indicate that loss of microRNA-16 may favor glioma angiogenesis, on the contrary overexpression of microRNA-16 in GBM cells plays a critical role in repressing endothelial function and angiogenesis by targeting Bmi-1. microRNA-16 may be a potential therapeutic agent in the treatment of GBM.