Revealing hidden patterns in deep neural network feature space continuum via manifold learning.

Deep neural networks (DNNs) extract thousands to millions of task-specific features during model training for inference and decision-making. While visualizing these features is critical for comprehending the learning process and improving the performance of the DNNs, existing visualization techniques work only for classification tasks. For regressions, the feature points lie on a high dimensional continuum having an inherently complex shape, making a meaningful visualization of the features intractable. Given that the majority of deep learning applications are regression-oriented, developing a conceptual framework and computational method to reliably visualize the regression features is of great significance. Here, we introduce a manifold discovery and analysis (MDA) method for DNN feature visualization, which involves learning the manifold topology associated with the output and target labels of a DNN. MDA leverages the acquired topological information to preserve the local geometry of the feature space manifold and provides insightful visualizations of the DNN features, highlighting the appropriateness, generalizability, and adversarial robustness of a DNN. The performance and advantages of the MDA approach compared to the existing methods are demonstrated in different deep learning applications.

Leveraging data-driven self-consistency for high-fidelity gene expression recovery.

Single cell RNA sequencing is a promising technique to determine the states of individual cells and classify novel cell subtypes. In current sequence data analysis, however, genes with low expressions are omitted, which leads to inaccurate gene counts and hinders downstream analysis. Recovering these omitted expression values presents a challenge because of the large size of the data. Here, we introduce a data-driven gene expression recovery framework, referred to as self-consistent expression recovery machine (SERM), to impute the missing expressions. Using a neural network, the technique first learns the underlying data distribution from a subset of the noisy data. It then recovers the overall expression data by imposing a self-consistency on the expression matrix, thus ensuring that the expression levels are similarly distributed in different parts of the matrix. We show that SERM improves the accuracy of gene imputation with orders of magnitude enhancement in computational efficiency in comparison to the state-of-the-art imputation techniques.

DNAJA1- and conformational mutant p53-dependent inhibition of cancer cell migration by a novel compound identified through a virtual screen.

Cancers are frequently addicted to oncogenic missense mutant p53 (mutp53). DNAJA1, a member of heat shock protein 40 (HSP40), also known as J-domain proteins (JDPs), plays a crucial role in the stabilization and oncogenic activity of misfolded or conformational mutp53 by binding to and preventing mutp53 from proteasomal degradation. However, strategies to deplete mutp53 are not well-established, and no HSP40/JDPs inhibitors are clinically available. To identify compounds that bind to DNAJA1 and induce mutp53 degradation, we performed an in silico docking study of ~10 million of compounds from the ZINC database for the J-domain of DNAJA1. A compound 7-3 was identified, and its analogue A11 effectively reduced the levels of DNAJA1 and conformational mutp53 with minimal effects on the levels of wild-type p53 and DNA-contact mutp53. A11 suppressed migration and filopodia formation in a manner dependent on DNAJA1 and conformational mutp53. A mutant DNAJA1 with alanine mutations at predicted amino acids (tyrosine 7, lysine 44, and glutamine 47) failed to bind to A11. Cells expressing the mutant DNAJA1 became insensitive to A11-mediated depletion of DNAJA1 and mutp53 as well as A11-mediated inhibition of cell migration. Thus, A11 is the first HSP40/JDP inhibitor that has not been previously characterized for depleting DNAJA1 and subsequently conformational mutp53, leading to inhibition of cancer cell migration. A11 can be exploited for a novel treatment against cancers expressing conformational mutp53.

Deep Learning-Based Water-Fat Separation from Dual-Echo Chemical Shift-Encoded Imaging.

Conventional water-fat separation approaches suffer long computational times and are prone to water/fat swaps. To solve these problems, we propose a deep learning-based dual-echo water-fat separation method. With IRB approval, raw data from 68 pediatric clinically indicated dual echo scans were analyzed, corresponding to 19382 contrast-enhanced images. A densely connected hierarchical convolutional network was constructed, in which dual-echo images and corresponding echo times were used as input and water/fat images obtained using the projected power method were regarded as references. Models were trained and tested using knee images with 8-fold cross validation and validated on out-of-distribution data from the ankle, foot, and arm. Using the proposed method, the average computational time for a volumetric dataset with ~400 slices was reduced from 10 min to under one minute. High fidelity was achieved (correlation coefficient of 0.9969, l1 error of 0.0381, SSIM of 0.9740, pSNR of 58.6876) and water/fat swaps were mitigated. I is of particular interest that metal artifacts were substantially reduced, even when the training set contained no images with metallic implants. Using the models trained with only contrast-enhanced images, water/fat images were predicted from non-contrast-enhanced images with high fidelity. The proposed water-fat separation method has been demonstrated to be fast, robust, and has the added capability to compensate for metal artifacts.

Implicit neural representation for radiation therapy dose distribution.

Objective. Dose distribution data plays a pivotal role in radiotherapy treatment planning. The data is typically represented using voxel grids, and its size ranges from 106to 108. A concise representation of the treatment plan is of great value in facilitating treatment planning and downstream applications. This work aims to develop an implicit neural representation of 3D dose distribution data.Approach. Instead of storing the dose values at each voxel, in the proposed approach, the weights of a multilayer perceptron (MLP) are employed to characterize the dosimetric data for plan representation and subsequent applications. We train a coordinate-based MLP with sinusoidal activations to map the voxel spatial coordinates to the corresponding dose values. We identify the best architecture for a given parameter budget and use that to train a model for each patient. The trained MLP is evaluated at each voxel location to reconstruct the dose distribution. We perform extensive experiments on dose distributions of prostate, spine, and head and neck tumor cases to evaluate the quality of the proposed representation. We also study the change in representation quality by varying model size and activation function.Main results. Using coordinate-based MLPs with sinusoidal activations, we can learn implicit representations that achieve a mean-squared error of 10-6and peak signal-to-noise ratio greater than 50 dB at a target bitrate of ∼1 across all the datasets, with a compression ratio of ∼32. Our results also show that model sizes with a bitrate of 1-2 achieve optimal accuracy. For smaller bitrates, performance starts to drop significantly.Significance. The proposed model provides a low-dimensional, implicit, and continuous representation of 3D dose data. In summary, given a dose distribution, we systematically show how to find a compact model to fit the data accurately. This study lays the groundwork for future applications of neural representations of dose data in radiation oncology.

Metal artifact reduction in 2D CT images with self-supervised cross-domain learning.

The presence of metallic implants often introduces severe metal artifacts in the x-ray computed tomography (CT) images, which could adversely influence clinical diagnosis or dose calculation in radiation therapy. In this work, we present a novel deep-learning-based approach for metal artifact reduction (MAR). In order to alleviate the need for anatomically identical CT image pairs (i.e. metal artifact-corrupted CT image and metal artifact-free CT image) for network learning, we propose a self-supervised cross-domain learning framework. Specifically, we train a neural network to restore the metal trace region values in the given metal-free sinogram, where the metal trace is identified by the forward projection of metal masks. We then design a novel filtered backward projection (FBP) reconstruction loss to encourage the network to generate more perfect completion results and a residual-learning-based image refinement module to reduce the secondary artifacts in the reconstructed CT images. To preserve the fine structure details and fidelity of the final MAR image, instead of directly adopting convolutional neural network (CNN)-refined images as output, we incorporate the metal trace replacement into our framework and replace the metal-affected projections of the original sinogram with the prior sinogram generated by the forward projection of the CNN output. We then use the FBP algorithms for final MAR image reconstruction. We conduct an extensive evaluation on simulated and real artifact data to show the effectiveness of our design. Our method produces superior MAR results and outperforms other compelling methods. We also demonstrate the potential of our framework for other organ sites.

Deep Neural Network With Consistency Regularization of Multi-Output Channels for Improved Tumor Detection and Delineation.

Deep learning is becoming an indispensable tool for imaging applications, such as image segmentation, classification, and detection. In this work, we reformulate a standard deep learning problem into a new neural network architecture with multi-output channels, which reflects different facets of the objective, and apply the deep neural network to improve the performance of image segmentation. By adding one or more interrelated auxiliary-output channels, we impose an effective consistency regularization for the main task of pixelated classification (i.e., image segmentation). Specifically, multi-output-channel consistency regularization is realized by residual learning via additive paths that connect main-output channel and auxiliary-output channels in the network. The method is evaluated on the detection and delineation of lung and liver tumors with public data. The results clearly show that multi-output-channel consistency implemented by residual learning improves the standard deep neural network. The proposed framework is quite broad and should find widespread applications in various deep learning problems.

A rapid screening classifier for diagnosing COVID-19.

Rationale: Coronavirus disease 2019 (COVID-19) has caused a global pandemic. A classifier combining chest X-ray (CXR) with clinical features may serve as a rapid screening approach. Methods: The study included 512 patients with COVID-19 and 106 with influenza A/B pneumonia. A deep neural network (DNN) was applied, and deep features derived from CXR and clinical findings formed fused features for diagnosis prediction. Results: The clinical features of COVID-19 and influenza showed different patterns. Patients with COVID-19 experienced less fever, more diarrhea, and more salient hypercoagulability. Classifiers constructed using the clinical features or CXR had an area under the receiver operating curve (AUC) of 0.909 and 0.919, respectively. The diagnostic efficacy of the classifier combining the clinical features and CXR was dramatically improved and the AUC was 0.952 with 91.5% sensitivity and 81.2% specificity. Moreover, combined classifier was functional in both severe and non-serve COVID-19, with an AUC of 0.971 with 96.9% sensitivity in non-severe cases, which was on par with the computed tomography (CT)-based classifier, but had relatively inferior efficacy in severe cases compared to CT. In extension, we performed a reader study involving three experienced pulmonary physicians, artificial intelligence (AI) system demonstrated superiority in turn-around time and diagnostic accuracy compared with experienced pulmonary physicians. Conclusions: The classifier constructed using clinical and CXR features is efficient, economical, and radiation safe for distinguishing COVID-19 from influenza A/B pneumonia, serving as an ideal rapid screening tool during the COVID-19 pandemic.

Deep Learning Prediction of Cancer Prevalence from Satellite Imagery.

The worldwide growth of cancer incidence can be explained in part by changes in the prevalence and distribution of risk factors. There are geographical gaps in the estimates of cancer prevalence, which could be filled with innovative methods. We used deep learning (DL) features extracted from satellite images to predict cancer prevalence at the census tract level in seven cities in the United States. We trained the model using detailed cancer prevalence estimates from 2018 available in the CDC (Center for Disease Control) 500 Cities project. Data from 3500 census tracts covering 14,483,366 inhabitants were included. Features were extracted from 170,210 satellite images with deep learning. This method explained up to 64.37% (median = 43.53%) of the variation of cancer prevalence. Satellite features are highly correlated with individual socioeconomic and health measures that are linked to cancer prevalence (age, smoking and drinking status, and obesity). A higher similarity between two environments is associated with better generalization of the model (p = 1.10-6). This method can be used to accurately estimate cancer prevalence at a high spatial resolution without using surveys at a fraction of the cost.

Machine learning techniques for biomedical image segmentation: An overview of technical aspects and introduction to state-of-art applications.

In recent years, significant progress has been made in developing more accurate and efficient machine learning algorithms for segmentation of medical and natural images. In this review article, we highlight the imperative role of machine learning algorithms in enabling efficient and accurate segmentation in the field of medical imaging. We specifically focus on several key studies pertaining to the application of machine learning methods to biomedical image segmentation. We review classical machine learning algorithms such as Markov random fields, k-means clustering, random forest, etc. Although such classical learning models are often less accurate compared to the deep-learning techniques, they are often more sample efficient and have a less complex structure. We also review different deep-learning architectures, such as the artificial neural networks (ANNs), the convolutional neural networks (CNNs), and the recurrent neural networks (RNNs), and present the segmentation results attained by those learning models that were published in the past 3 yr. We highlight the successes and limitations of each machine learning paradigm. In addition, we discuss several challenges related to the training of different machine learning models, and we present some heuristics to address those challenges.

CD4+ αß T cell infiltration into the leptomeninges of lumbar dorsal roots contributes to the transition from acute to chronic mechanical allodynia after adult rat tibial nerve injuries.

BACKGROUND: Antigen-specific and MHCII-restricted CD4+ αß T cells have been shown or suggested to play an important role in the transition from acute to chronic mechanical allodynia after peripheral nerve injuries. However, it is still largely unknown where these T cells infiltrate along the somatosensory pathways transmitting mechanical allodynia to initiate the development of chronic mechanical allodynia after nerve injuries. Therefore, the purpose of this study was to ascertain the definite neuroimmune interface for these T cells to initiate the development of chronic mechanical allodynia after peripheral nerve injuries. METHODS: First, we utilized both chromogenic and fluorescent immunohistochemistry (IHC) to map αß T cells along the somatosensory pathways for the transmission of mechanical allodynia after modified spared nerve injuries (mSNIs), i.e., tibial nerve injuries, in adult male Sprague-Dawley rats. We further characterized the molecular identity of these αß T cells selectively infiltrating into the leptomeninges of L4 dorsal roots (DRs). Second, we identified the specific origins in lumbar lymph nodes (LLNs) for CD4+ αß T cells selectively present in the leptomeninges of L4 DRs by two experiments: (1) chromogenic IHC in these lymph nodes for CD4+ αß T cell responses after mSNIs and (2) fluorescent IHC for temporal dynamics of CD4+ αß T cell infiltration into the L4 DR leptomeninges after mSNIs in prior lymphadenectomized or sham-operated animals to LLNs. Finally, following mSNIs, we evaluated the effects of region-specific targeting of these T cells through prior lymphadenectomy to LLNs and chronic intrathecal application of the suppressive anti-αßTCR antibodies on the development of mechanical allodynia by von Frey hair test and spinal glial or neuronal activation by fluorescent IHC. RESULTS: Our results showed that during the sub-acute phase after mSNIs, αß T cells selectively infiltrate into the leptomeninges of the lumbar DRs along the somatosensory pathways responsible for transmitting mechanical allodynia. Almost all these αß T cells are CD4 positive. Moreover, the temporal dynamics of CD4+ αß T cell infiltration into the lumbar DR leptomeninges are specifically determined by LLNs after mSNIs. Prior lymphadenectomy to LLNs specifically reduces the development of mSNI-induced chronic mechanical allodynia. More importantly, intrathecal application of the suppressive anti-αßTCR antibodies reduces the development of mSNI-induced chronic mechanical allodynia. In addition, prior lymphadenectomy to LLNs attenuates mSNI-induced spinal activation of glial cells and PKCγ+ excitatory interneurons. CONCLUSIONS: The noteworthy results here provide the first evidence that CD4+ αß T cells selectively infiltrate into the DR leptomeninges of the somatosensory pathways transmitting mechanical allodynia and contribute to the transition from acute to chronic mechanical allodynia after peripheral nerve injuries.

ProBDNF inhibits collective migration and chemotaxis of rat Schwann cells.

Schwann cell migration, including collective migration and chemotaxis, is essential for the formation of coordinate interactions between Schwann cells and axons during peripheral nerve development and regeneration. Moreover, limited migration of Schwann cells imposed a serious obstacle on Schwann cell-astrocytes intermingling and spinal cord repair after Schwann cell transplantation into injured spinal cords. Recent studies have shown that mature brain-derived neurotrophic factor, a member of the neurotrophin family, inhibits Schwann cell migration. The precursor form of brain-derived neurotrophic factor, proBDNF, was expressed in the developing or degenerating peripheral nerves and the injured spinal cords. Since "the yin and yang of neurotrophin action" has been established as a common sense, proBDNF would be expected to promote Schwann cell migration. However, we found, in the present study, that exogenous proBDNF also inhibited in vitro collective migration and chemotaxis of RSC 96 cells, a spontaneously immortalized rat Schwann cell line. Moreover, proBDNF suppressed adhesion and spreading of those cells. At molecular level, proBDNF inhibits F-actin polymerization and focal adhesion dynamics in cultured RSC 96 cells. Therefore, our results suggested a special case against the classical opinion of "the yin and yang of neurotrophin action" and implied that proBDNF might modulate peripheral nerve development or regeneration and spinal cord repair through perturbing native or transplanted Schwann cell migration.

Behavioral characterization of neuropathic pain on the glabrous skin areas reinnervated solely by axotomy-regenerative axons after adult rat sciatic nerve crush.

In cranial and spinal nerve ganglia, both axotomized primary sensory neurons without regeneration (axotomy-nonregenerative neurons) and spared intact primary sensory neurons adjacent to axotomized neurons (axotomy-spared neurons) have been definitely shown to participate in pain transmission in peripheral neuropathic pain states. However, whether axotomized primary sensory neurons with regeneration (axotomy-regenerative neurons) would be integral components of neural circuits underlying peripheral neuropathic pain states remains controversial. In the present study, we utilized an adult rat sciatic nerve crush model to systematically analyze pain behaviors on the glabrous plantar surface of the hindpaw sural nerve skin territories. To the best of our knowledge, our results for the first time showed that heat hyperalgesia, cold allodynia, mechanical allodynia, and mechanical hyperalgesia emerged and persisted on the glabrous sural nerve skin areas after adult rat sciatic nerve crush. Interestingly, mechanical hyperalgesia was sexually dimorphic. Moreover, with our optimized immunofluorescence staining protocol of free-floating thick skin sections for wide-field epifluorescence microscopic imaging, changes in purely regenerative reinnervation on the same skin areas by axotomized primary sensory afferents were shown to be paralleled by those pathological pain behaviors. To our surprise, Protein Gene Product 9.5-immunoreactive nerve fibers with regular and large varicosities ectopically emigrated into the upper dermis of the glabrous sural nerve skin territories after adult rat sciatic nerve crush. Our results indicated that axotomy-regenerative primary sensory neurons could be critical elements in neural circuits underlying peripheral neuropathic pain states. Besides, our results implied that peripheral neuropathic pain transmitted by axotomy-regenerative primary sensory neurons alone might be a new dimension in the clinical therapy of peripheral nerve trauma beyond regeneration.

[Early Systemic Administration of IL-10 Inhibits Neuropathic Pain in Adult Rats with Tibial Nerve Injury].

OBJECTIVES: To determine the effect of early systemic administration of IL-10 on peripheral neuropathic pain induced by tibial nerve permanent transection [modified spared nerve injury (mSNI)]in adult rats. METHODS: Male adult Sprague-Dawley (SD) rats (ten-week old, 250-300 g) with mSNI were randomly divided into mSNI, sham-operated, IL-10 intervention (intraperitoneal injection), PBS intervention (intraperitoneal injection) groups, each containing six rats. Intraperitoneally injections (IL-10 or PBS) were given immediately after surgeries for a single regime with a dosage of 500 uL (0.1 mg/mL). Plantar test, von Frey hairs test, pinprick test and acetone test were performed before and after tibial nerve injuries (0 d, 4/5 d, 7/8 d, 14/15 d) to evaluate region-specific pain responses of the rats on the plantar sural and saphenous skin territories of ipsilateral and contralateral hindpaws. The hindpaw position (on 8 d) of six additional rats with standard SNI was compared with those with mSNI. RESULTS: The rats with standard SNI showed an eversion posture of hindpaws, more prominent than those with mSNI. Region-specific pathological pain evoked by mechanical and thermal stimuli on the sural and saphenous skin territories of the plantar surfaces of rat hindpaws was demonstrated on the ipsilateral rather than contralateral hindpaws. This effect was shown in the rats with mSNI but not in those with sham operations. Compared with PBS, early intraperitoneal injection of IL-10 significantly and persistently attenuated either allodynia or hyperalgesia in the rats with mSNI. CONCLUSIONS: Tibial nerve permanent transection models of adult rats can be used as a simple but useful rodent model of peripheral neuropathic pain. Early systemic administration of IL-10 impairs the pathogenesis of neuropathic pain induced by tibial nerve injuries.

[Role of focal adhesion kinase in adhesion and migration of Hep G2 cells].

Focal adhesion kinase (FAK) plays a critical role in the process of cell adhesion and migration by regulating the expression of downstream small G proteins. A kind of focal adhesion kinase (FAK) inhibitor was used to inhibit the phosphorylation of Y397 site of FAK, and scratch wound migration assay was used to examine the effect of FAK inhibitor with different concentrations (0-250 nmol/mL) on the migration of hepatomal cells (Hep G2 cells) at 0, 2, 4, 8 and 24h. Immunofluorescence analysis and Western blot analysis were performed to detect the expression of F-actin and small G proteins Rac1, RhoA and Cdc42 in Hep G2 cells treated with FAK inhibitor for 120 min. The results indicated that the FAK inhibitor can inhibit the migration of Hep G2 cells with a dose- and time-dependent manner. F-actin was down-regulated in Hep G2 cells treated with FAK inhibitor for 120 min, and expression of small G proteins were inhibited at different durations. The inhibition of FAK phosphorylation could inhibit cell adhesion and migration by down-regulating small G proteins. These results suggested that FAK inhibitor can inhibit the migration of tumor cells by blocking FAK phosphorylation. This means that FAK inhibitor can block the metastasis of tumor cells to surrounding tissues. It may be a potential application in the prevention and treatment of cancer.

[Study on FAK regulation of migration of vascular endothelial cells depending upon focal adhesion proteins].

Tumor angiogenesis induced by vascular endothelial cells (VECs) migration is a necessary condition for tumor growth and metastasis. The purpose of this study is to investigate the effect of focal adhesion kinase (FAK) inhibitor (50nmol/mL) on the adhesion and migration of endothelial cells(ECs) and the expression of focal adhesion proteins vinculin, talin and paxillin. Scratch wound migration assay was performed to examine the effect of FAK inhibitor with 50nmol/mL on ECs migration at 0, 5, 10, 30, 60 and 120min, respectively. And immunofluorescence analysis was performed to detect the expression of F-actin in ECs treated with FAK inhibitor within 2h. Western blot was carried out to determine the effect of FAK inhibitor on expression of vinculin, talin and paxillin proteins. The results showed that the migration distance and the expression of F-actin in ECs treated with FAK inhibitor decreased significantly compared with that of the controls, and the level of vinculin showed no significant difference with increasing of treated time of FAK inhibitor. However, the talin and paxillin showed an identical decreasing tendency in 5-10min, but slowly going up in 30min and then after subsequently decreasing. The results of this study proved that blocking phosphorylation of FAK could inhibit VECs adhesion and migration by downregulating focal adhesion proteins so that it may inhibit tumor angiogenesis. This may provide a new approach for tumor therapy.

[Influence of titanium particles loading on releases of cytokines of osteoblasts].

Aseptic loosening is mainly mediated by bone resorption cytokines in the surrounding area of orthopaedic implants. Our previous investigation demonstrtated that different-sized titanium particles loading can inbibit the osteoblastic differentiation and mineraliztion. In order to investigate the hypothesis that particulate wear debris derived from prosthetic biomaterials affects the release of bone resorption related cytokines, we studied the influence of different-sized titanium particles loading on the osteoblastic cytokines by assaying the secretion of IL-6, IL-10 with use of ABC-quantitative sandwich enzyme-linked immunosorbent assay, and on the expression of osteoclast differentiation factors (ODF) by RT-PCR. The results showed that the 0.9 microm titanium particles promoted osteoblasts producing bone resorption cytokines (IL-6, ODF), and simultaneously secreted bone absorption restraining factor (IL-10) quickly and transitorily. In comparison, the 2.7 microm and 6.9 microm titanium particles,especially the latter primarily promoted osteoblasts secreting bone absorption promoting factors powerfully and slowly. The results suggested that there was a biphasic response appearing in titanium particles loaded-osteoblastic cultures, the level of which varied according to the different size and the loading time of titanium particles. This in vitro experimental result showed that attentaion to the inhibition of bone resorption cytokines stimulated by wear debris and to the screen potential favourable biomaterials for implants must be taken.