MB Model Research Articles

Enabling Intelligence on the Edge: Leveraging Edge Impulse to Deploy Multiple Deep Learning Models on Edge Devices for Tomato Leaf Disease Detection

Tomato diseases, including Leaf blight, Leaf curl, Septoria leaf spot, and Verticillium wilt, are responsible for up to 50% of annual yield loss, significantly impacting global tomato production, valued at approximately USD 87 billion. In Ghana, there is a yield gap of about 50% in tomato production, which requires drastic measures to increase the yield of tomatoes. Conventional diagnostic methods are labor-intensive and impractical for real-time application, highlighting the need for innovative solutions. This study addresses these issues in Ghana by utilizing Edge Impulse to deploy multiple deep-learning models on a single mobile device, facilitating the rapid and precise detection of tomato leaf diseases in the field. This work compiled and rigorously prepared a comprehensive Ghanaian dataset of tomato leaf images, applying advanced preprocessing and augmentation techniques to enhance robustness. Using TensorFlow, we designed and optimized efficient convolutional neural network (CNN) architectures, including MobileNet, Inception, ShuffleNet, Squeezenet, EfficientNet, and a custom Deep Neural Network (DNN). The models were converted to TensorFlow Lite format and quantized to int8, substantially reducing the model size and improving inference speed. Deployment files were generated, and the Edge Impulse platform was configured to enable multiple model deployments on a mobile device. Performance evaluations across edge hardware provided metrics such as inference speed, accuracy, and resource utilization, demonstrating reliable real-time detection. EfficientNet achieved a high training accuracy of 97.12% with a compact 4.60 MB model size, proving its efficacy for mobile device deployment. In contrast, the custom DNN model is optimized for microcontroller unit (MCU) deployment. This edge artificial intelligence (AI) technology integration into agricultural practices offers scalable, cost-effective, and accessible solutions for disease classification, enhancing crop management, and supporting sustainable farming practices.

YOLOSAR-Lite: a lightweight framework for real-time ship detection in SAR imagery

ABSTRACT Real-time ship detection using SAR images is crucial for maritime operations but remains challenging due to issues such as sidelobes, complex background interferences, and target defocusing. Studies have shown that deep learning can address these problems, current deep learning algorithms often suffer from achieving the necessary accuracy due to complex noise and background characteristics. Additionally, advanced algorithms excelling in feature extraction usually require significant computational resources. This paper proposes a novel deep learning approach based on YOLOv8 to improve accuracy while reducing model complexity and computational burden. The approach innovatively integrates a knowledge distillation module into the YOLOv8. We further enhance the model by replacing the original YOLOv8 backbone, neck, and head with lightweight alternatives: HGNetv2, SlimNeck, and a newly designed decoupled head. The enhanced model, YOLOSAR-Lite, demonstrated exceptional performance on both accuracy and model complexity metrics. Rigorous evaluation on the Official-SSDD dataset and the SAR-Ship-Dataset revealed recognition accuracies of 95.32% and 96.06%, respectively. Furthermore, YOLOSAR-Lite achieved significant reduction in complexity compared to the YOLOv8: 4.48G FLOPs (45.30% reduction), 2.05M parameters (31.89% reduction), and 3.91MB model size (31.76% reduction). These improvements demonstrate YOLOSAR-Lite achieves higher accuracy with minimal complexity. Consequently, YOLOSAR-Lite is a promising solution for real-time ship detection.

Efficient face anti-spoofing via head-aware transformer based knowledge distillation with 5 MB model parameters

Although face recognition technology has been applied in many scenarios, it still suffers from many types of presentation attacks, so face anti-spoofing (FAS) becomes a hot topic in computer vision. Recently, vision transformer is recognized as the mainstream architecture for FAS, which always relies on auxiliary information, sophisticated tricks and huge model parameters. Considering that face based identity authentication usually takes place on mobile-like devices, therefore how to design an effective and lightweight model is of great significance. Inspired by the powerful global modeling ability of self-attention and the model compression ability of knowledge distillation, a simple yet effective knowledge distillation approach is proposed for FAS under transformer framework. Our primary idea is to leverage the rich knowledge of a teacher network pre-trained on large-scale face data to guide the learning of a lightweight student network. The main contributions of our method are threefold: (1) Feature- and logits-level distillation are combined to transfer the rich knowledge of teacher to student. (2) A head-aware strategy is proposed to deal with the dimension mismatching issue of middle encoder layers between teacher and student networks, in which a novel attention head correlation matrix is introduced. (3) Our method can bridge the performance gap between teacher and student, and the resulting student network is extremely lightweight with only 5 MB parameters. Extensive experiments are conducted on three public face-spoofing datasets, CASIA-FASD, Replay-Attack and OULU-NPU, the results demonstrate that our method can obtain performance on par with or superior to most FAS methods and outperform many knowledge distillation methods. Meanwhile, the distilled student network achieves excellent performance with 17× fewer parameters and 9× faster inference time compared to the teacher network. The code will be publicly available at https://github.com/Maricle-zhangjun/HaTFAS.

Detection of various gastrointestinal tract diseases through a deep learning method with ensemble ELM and explainable AI

The rising prevalence of gastrointestinal (GI) tract disorders worldwide highlights the urgent need for precise diagnosis, as these diseases greatly affect human life and contribute to high mortality rates. Fast identification, accurate classification, and efficient treatment approaches are essential for addressing this critical health issue. Common side effects include abdominal pain, bloating, and discomfort, which can be chronic and debilitating. Nausea and vomiting are also frequent, leading to difficulties in maintaining adequate nutrition and hydration. The current study intends to develop a deep learning (DL)-based approach that automatically classifies GI tract diseases. For the first time, a GastroVision dataset with 8000 images of 27 different GI diseases was utilized in this work to design a computer-aided diagnosis (CAD) system. This study presents a novel lightweight feature extractor with a compact size and minimum number of layers named Parallel Depthwise Separable Convolutional Neural Network (PD-CNN) and a Pearson Correlation Coefficient (PCC) as the feature selector. Furthermore, a robust classifier named the Ensemble Extreme Learning Machine (EELM), combined with pseudo inverse ELM (ELM) and L1 Regularized ELM (RELM), has been proposed to identify diseases more precisely. A hybrid preprocessing technique, including scaling, normalization, and image enhancement techniques such as erosion, CLAHE, sharpening, and Gaussian filtering, are employed to enhance image representation and improve classification performance. The proposed approach consists of twenty-four layers and only 0.815 million parameters with a 9.79 MB model size. The proposed PD-CNN-PCC-EELM extracts essential features, reduces computational overhead, and achieves excellent classification performance on multiclass GI images. The PD-CNN-PCC-EELM achieved the highest precision, recall, f1, accuracy, ROC-AUC, and AUC-PR values of 88.12 ± 0.332 %, 87.75 ± 0.348 %, 87.12 ± 0.324 %, 87.75 %, 98.89 %, and 92 %, respectively, while maintaining a minimum testing time of 0.000001 s. A comparative study utilizes 10-fold cross-validation, ablation study and various state-of-the-art (SOTA) transfer learning (TL) models as feature extractors. Then, the PCC and EELM are integrated with TL to generate predictions, notably in terms of performance and real-time processing capability; the proposed model significantly outperforms the other models. Moreover, various explainable AI (XAI) methods, such as SHAP (Shapley Additive Explanations), heatmap, guided heatmap, Grad-Cam (Gradient-weighted Class Activation Mapping), guided Grad-CAM, and guided Saliency mapping, have been employed to explore the interpretability and decision-making capability of the proposed model. Therefore, the model provides practical intelligence for increasing confidence in diagnosing GI diseases in real-world scenarios.

MODL-06. USING HYDROGEL MODELS TO RECAPITULATE THE TUMOUR-IMMUNE MICROENVIRONMENT OF AGGRESSIVE GROUP 4 MEDULLOBLASTOMA TUMOURS

Abstract BACKGROUND Medulloblastoma is the most common malignant childhood brain-tumour, predominantly affecting children between the ages of one and nine. It is divided into four molecular subgroups amongst which, Group 3 (G3) and Group 4 (G4) subgroups have the worst prognosis. Subgroups possess distinct extracellular-matrix (ECM) and tumour immune microenvironments (TIME) directly influencing therapy-response and metastasis. Current models of medulloblastoma fail to precisely recreate the complex, subgroup-specific TIME and do not mimic intricate immune-tumour heterotypic signalling facilitating immune-evasion. Here we use a HyStem-hydrogel which forms a hyaluronan-rich environment mimicking the natural brain architecture. This overcomes several limitations of current MB models, allowing customisation of subgroup matching of ECM components, adjustable ECM-stiffness and co-culturing of immune cells. METHODS HyStem-hydrogels were co-assembled with ECM components (with/without laminin) and G4 medulloblastoma cell line CHLA-01-MED. Hydrogels were imaged every three-days using an EVOS M7000 system and assessed for cell-viability with PrestoBlue every seven days. Co-culture laminin-supplemented HyStem hydrogels were established with M1-like or M2-like resting/activated macrophages and their growth tracked in the same manner. On day twenty-eight, hydrogels were fixed and stained for markers of cellular proliferation (Ki-67), ECM (laminin, vitronectin) and macrophage markers. Media from hydrogel co-cultures was collected and ELISAs permofmed to assess cytokine production. RESULTS Hydrogels with/without laminin both remained viable and stable for 28 days. Laminin-supplemented hydrogels displayed a mixed laminar-nodular phenotype which mirrored the heterogenous nature of G4 patient-tumours. Co-culture of CHLA-01-MED cells with activated-macrophages led to cytokine production and altered growth patterns. CONCLUSION We show that HyStem hydrogels recapitulate biologically relevant behaviour of aggressive G4-MB tumours in vitro and represent a promising model for G4-tumours. Future research will involve inclusion of additional brain relevant cells/ECM components and comparisons to a fully customisable peptide-amphiphile hydrogel.

Abstract 5107: Targeting CTPS1 and nucleotide biosynthesis pathways as a novel therapeutic approach in MYC-amplified medulloblastoma

Abstract Although leukemia is the leading pediatric oncology diagnosis, brain tumors are the leading cause of pediatric cancer-related death. Medulloblastoma is the most common pediatric brain tumor, representing approximately 20% of all diagnoses. Of all four defined subgroups, MYC-amplified Group 3 medulloblastoma (G3 MB) tumors are the deadliest, with a < 50% 5-year overall survival rate. High risk patients receive large doses of chemo- and radio- therapy (CT/RT) with most survivors experiencing debilitating neurological and cognitive sequelae making the identification of novel targets of the utmost importance. Utilizing whole-genome CRISPR screening from the Pediatric Cancer Dependency Map, we identified the enzyme CTPS1 as a strong G3 MB-specific vulnerability. Recent studies corroborate our findings and suggest a dependence on the de novo pyrimidine biosynthesis pathway in MB. Targeted knockdown of CTPS1 restricts sphere formation and reduces proliferation of G3 MB tumor lines. We confirmed this genetic approach using a specific CTPS1 inhibitor, which also reduced proliferation but did not induce cell death, suggesting a cytostatic mechanism. Strikingly, we demonstrate that G3 MB is only dependent upon de novo pyrimidine synthesis and not de novo purine synthesis, despite regulation of both pathways by oncogenic MYC signaling. Previous work has shown that triggering an imbalance in pyrimidine and purine levels leads to replication stress and activation of the ATR/CHK1 pathway. Both genetic and pharmacologic inhibition of CTPS1 leads to CHK phosphorylation, allowing for cell survival despite lack of dNTPs. Previously, the CHK1 inhibitor, prexasertib, demonstrated robust brain accumulation and targeting of G3 MB in pre-clinical models. Dual inhibition of CHK1 and CTPS1 synergizes to significantly reduce cell viability in vitro and tumor growth in vivo. We further demonstrated robust synergy of CHK1 inhibition with either glutamine antagonists or other inhibitors of the de novo pyrimidine synthesis pathway, prolonging in vivo survival in an orthotopic xenograft G3 MB model. The results of this investigation confirm CTPS1 as a novel target in G3 MB, pinpoint a dependency on de novo pyrimidine rather than purine biosynthesis, and identifies a novel rational combination for treatment in MYC-amplified MB patients. Citation Format: Matthew Hathaway, Hawa Jagana, Michael Meechan, John Hemenway, Kate E. Gadek, Aya Miyaki, Isabella Terrones, Barbara Slusher, Siobhan Pattwell, Nicholas Vitanza, Myron K. Evans. Targeting CTPS1 and nucleotide biosynthesis pathways as a novel therapeutic approach in MYC-amplified medulloblastoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 5107.

Abstract 5111: Rational synergistic radiosensitization strategies for group 3 medulloblastoma

Abstract Medulloblastoma (MB) is the most common malignant pediatric brain tumor. MB is comprised of 4 groups, termed Wnt, Shh, Group 3 and Group 4, each characterized by distinct molecular and clinical profiles. MYC-amplified Group 3 MB is associated imminent fatal metastatic relapse following standard treatment approaches, which currently include surgery, craniospinal irradiation and chemotherapy, highlighting an urgent need for identification of therapeutic vulnerabilities for these tumours. As such, using 4 novel patient-derived models of Group 3 MB that faithfully recapitulate the molecular and histological tumor characteristics, we characterised response to radiation of Group 3 MB in vitro to show that these tumors do not possess intrinsic radiation resistance. Furthermore, cumulative radiation experiments do not increase bulk radiation resistance in MB cells, highlighting the contribution of dynamic cell states to radiation response. In contrast, exposure to radiation leads to changes in cell clonogenic capacity as measured by limited dilution assays. Furthermore, RNA-seq data uncovered altered transcriptomic programs in response to cumulative radiation treatment. In particular, we observe differential regulation of genes encoding potassium ion channels and their regulators as well as genes encoding ECM and cell adhesion proteins. As part of our goal to evaluate the current multimodal treatment strategy of MB, we also show that a set of standard chemotherapeutic agents (etoposide, thiotepa and lomustine) currently used in the clinic do not produce a potent response at optimal doses in vitro, potentially accounting for abysmal survival outcomes seen in chemotherapy treated MB patients. As such, we conducted a screen of 144 FDA approved cancer drugs and identified therapeutics that selectively target Group 3 MB cells in contrast to normal tissue controls (foetal neural stem cells) in the low nanomolar dose range. In particular, topoisomerase inhibitor topotecan produces a potent response specific to MYC-amplified Group 3 MB, and sensitizes 2 patient-derived MB models to gamma irradiation. Altogether, our data characterizes Group 3 MB molecular and cellular responses to current treatment modalities and provides rationale for new clinical strategies to sensitize patients to radiation that will potentially allow to reduce the required radiation doses to treat these aggressive tumors, reduce the incidence of radiation treatment-related morbidity as well as improve cure rates. Citation Format: Julija Povilaikaite, Kaitlin Stitz, Nikhita Austin, Vijay Ramaswamy. Rational synergistic radiosensitization strategies for group 3 medulloblastoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 5111.

Acoustic emissions based estimation of the temporal changes in microbubble radius during ultrasonic excitation

Our ability to study microbubble dynamics in vivo and link them to distinct mechano-biological effects hinges on our ability to accurately estimate the temporal changes in MB radius during ultrasonic excitation. Here, we hypothesize that real-time passive cavitation detection (PCD) monitoring combined with linear acoustic wave propagation theory can accurately estimate stable MB radius dynamics under in vivo conditions. To test our hypothesis, we employed numerical simulations, based on Rayleigh–Plesset modeling, followed by experimental validation, using calibrated PCDs with concurrent optical imaging of MB dynamics using high frame rate microscopy. Our method termed linear acoustic wave propagation and superposition (LAWPS) algorithm, combines Fourier series expansion with Euler’s relationship to estimate the acoustic emission (AE) from single MB, which is considered as a monopole source (i.e., R0 + ΔR < λ). LAWPS algorithm, which is independent of MB properties and, thus, can be linearly reversed to calculate MB oscillation radius from AE, was able to accurately capture the radiated pressure generated from a Rayleigh–Plesset MB modeling. Crucially, inverse LAWPS algorithm onto AE generated by Vokurka et al. resulted in the original MB model. Experimental observation using monodisperse MBs was able to accurately estimate the temporal changes in MB radius during 0.5 MHz ultrasonic excitation

Efficient Small-Object Detection in Underwater Images Using the Enhanced YOLOv8 Network

Underwater object detection plays a significant role in marine ecosystem research and marine species conservation. The improvement of related technologies holds practical significance. Although existing object-detection algorithms have achieved an excellent performance on land, they are not satisfactory in underwater scenarios due to two limitations: the underwater objects are often small, densely distributed, and prone to occlusion characteristics, and underwater embedded devices have limited storage and computational capabilities. In this paper, we propose a high-precision, lightweight underwater detector specifically optimizing for underwater scenarios based on the You Only Look Once Version 8 (YOLOv8) model. Firstly, we replace the Darknet-53 backbone of YOLOv8s with FasterNet-T0, reducing model parameters by 22.52%, FLOPS by 23.59%, and model size by 22.73%, achieving model lightweighting. Secondly, we add a Prediction Head for Small Objects, increase the number of channels for high-resolution feature map detection heads, and decrease the number of channels for low-resolution feature map detection heads. This results in a 1.2% improvement in small-object detection accuracy, while the remaining model parameters and memory consumption are nearly unchanged. Thirdly, we use Deformable ConvNets and Coordinate Attention in the neck part to enhance the accuracy in the detection of irregularly shaped and densely occluded small targets. This is achieved by learning convolution offsets from feature maps and emphasizing the regions of interest (RoIs). Our method achieves 52.12% AP on the underwater dataset UTDAC2020, with only 8.5 M parameters, 25.5 B FLOPS, and 17 MB model size. It surpasses the performance of large model YOLOv8l, at 51.69% AP, with 43.6 M parameters, 164.8 B FLOPS, and 84 MB model size. Furthermore, by increasing the input image resolution to 1280 × 1280 pixels, our model achieves 53.18% AP, making it the state-of-the-art (SOTA) model for the UTDAC2020 underwater dataset. Additionally, we achieve 84.4% mAP on the Pascal VOC dataset, with a substantial reduction in model parameters compared to previous, well-established detectors. The experimental results demonstrate that our proposed lightweight method retains effectiveness on underwater datasets and can be generalized to common datasets.

YOLO_MRC: A fast and lightweight model for real-time detection and individual counting of Tephritidae pests

Tephritidae pests severely affect the quality and safety of various melons, fruits and vegetable crops. However, many agricultural managers lack an adequate understanding of the level of pest occurrence, resulting in the misuse of pesticides, which ultimately leads to environmental pollution and economic loss. Therefore, real-time detection and counting of Tephritidae pests are important for timely pest spotting and control. This work helps quickly determine the distribution and abundance of pests in the current environment, thus providing data on pest conditions for agricultural management to optimize pesticide use. Nevertheless, the fast speed, high accuracy, and lightweight performance of real-time detection and counting are difficult to balance. To address this problem, based on the YOLOv8n model, this paper takes Bactrocera cucurbitae pests as the detection target and proposes a fast and lightweight real-time detection and individual counting model for Tephritidae pests, named YOLO_MRC. This paper introduces three key improvements: (1) Constructing a new module called Multicat into the neck network increases the focus on the detection target by incorporating an attention mechanism; (2) Reducing the original three detection heads to two and then adjusting their sizes to decrease the number of parameters in the network model; (3) Devising a novel module, C2flite, to enhance the deep feature extraction capability of the backbone network. According to the above points, this paper conducts ablation experiments to compare the performances of different models. Experiments showed that the Multicat module significantly offsets the large increase in GFLOPs and processing time caused by reducing the detection head and can further reduce the number of parameters and improve the accuracy when combined with the C2flite module. On our Bactrocera cucurbitae pest dataset, the mAP@0.5 of the YOLO_MRC model reached 99.3%. Simultaneously, as the number of parameters decreases by 63.68%, GFLOPs is reduced by 19.75%, and the processing time is shortened by 5%. To ensure the validity of the model, YOLO_MRC is compared with four excellent detection models by using manual counting results as the benchmark. YOLO_MRC achieves an average pest counting accuracy of 94%, demonstrating superior performance in terms of model size and processing time. To further explore the performance of YOLO_MRC in multiclass insect detection tasks, we choose the public dataset Pest_24_640 for comparison with four state-of-the-art models. YOLO_MRC achieves a 3.6 ms processing time and 70.4% accuracy with only a 2.4 MB model size, which demonstrates the potential of YOLO_MRC in multiclass pest detection.

Predicting real-time within-vehicle air pollution exposure with mass-balance and machine learning approaches using on-road and air quality data

Modelling the air pollutant concentrations within-vehicles is an essential step to estimate our daily exposure to air pollution. This is a challenging issue however, since the processes that affect the exposures within-vehicles change with different driving patterns and ventilation settings. This study introduces an innovative approach that combines mass-balance principles and machine learning techniques, leveraging ambient air quality, on-road and within-vehicle measurements of particulate matter (PM10, PM2.5, PM1), nitrogen dioxide (NO2), nitrogen oxides (NOx), aerosol lung surface deposited area (LSDA) and ultrafine particles (UFP) under different ventilation settings to estimate air pollution exposure levels within vehicles. The first model (MB) includes basic physical and chemical processes and follows a mass-balance approach to estimate the within-vehicle concentrations. The second model (ML) applies data driven machine learning algorithms to a training set of observations to predict unseen within-vehicle concentrations. By using a number generator, the whole observational dataset was divided to 80:20 and 80% was used to build and train the ML model, while 20% was used for validation. Both models demonstrated good predictions of observations apart from an underestimation in UFP and LSDA. The ML model showed better predictive power than the MB model and had skill in predicting the unseen within-vehicle exposures. The ML model predictions were as good as the MB model for most of the species and improved for NO2. The ML model demonstrated good index of agreement (IOA >0.69) and Pearson correlation coefficient (r > 0.80) for all the species. The inclusion of air quality data from nearby monitoring stations instead of on-road (sampled while driving), in the ML model showed promising and new capabilities to within-vehicle exposure predictions. In an era where air pollution is a growing concern, understanding and predicting within-vehicle air pollution exposure is of great importance for public health and environmental research. This research not only advances the field of exposure assessment but (at no extra cost) also demonstrates practical implications for real-time exposure mapping and health impact assessment of vehicle occupants with existing infrastructure.

A site-site interaction two-dimensional model with water like structural properties

A site-site interaction model is proposed for water in two-dimension, as an alternative to the traditional Mercedes-Benz model. In MB model, water molecules are modelled as 2-dimensional Lennard-Jones disks with three hydrogen bonding arms arranged symmetrically, resembling the Mercedes-Benz logo. The MB model qualitatively predicts both the anomalous properties of pure water and the anomalous solvation thermodynamics of non-polar molecules. One of the features of this earlier model was to have a pair correlation function with first peak for the Lennard-Jones contact distinct of that corresponding to the hydrogen bonding, which is very different from real water which has a single first peak, but a dual peak for the structure factor. The site-site model proposed here reproduces this typical feature of real water, both in real and reciprocal space. It also reproduces several of the known anomalies of real water, such as the density maximum. In addition, because of the screened Coulomb interaction between the sites, the new model appear to exhibit more homogeneity that the MB models and their variants, the latter which is highlighted by a k=0 increase of their structure factors. The new model transfers the usual bond order paradigm into a charge order paradigm, enforcing atom-atom interactions over orientational interactions.

MDB-27. A PRO-DRUG OF TRIPTOLIDE IN CLINICAL DEVELOPMENT ATTENUATES HIGH-RISK MEDULLOBLASTOMA GROWTH

Abstract Medulloblastoma (MB), the most common malignant brain tumor of childhood, is currently classified in four major subgroups. A natural compound used in Chinese medicine for centuries, triptolide, has been shown to have a potent antitumor activity, and the efficacy of one of its derivatives is being evaluated in the clinic. However, to date no preclinical data supports the use of triptolide for the treatment of patients with MB. Here, we performed bioinformatic analyses to predict which subgroup of medulloblastoma patients would more likely benefit from the use of triptolide, and validated its efficacy in series of ex vivo and in vivo mouse and human derived MB models. Correlation analyses identified Group 3 (G3) MB patients as the most prone to respond to triptolide treatment. Consistent with this prediction, triptolide attenuated the growth of G3 MB at lower doses than those classified as SHH subgroup, increasing symptom-free survival in G3 mouse MB models. Transcriptomic analyses revealed MYC signaling as the key downstream target of triptolide in MB cells. Importantly, a pro-drug of triptolide in clinical development, minnelide, similarly showed efficacy in mouse and human derived G3 in vivo MB models. Our findings highlight the potential of repurposing Minnelide to treat G3 MB patients. Due to their dismal prognosis, inclusion of G3 MB patients in future minnelide clinical trials should be considered.

Wear Model of a Mechanical Seal Based on Piecewise Fractal Theory

In this research, a model is proposed using piecewise fractal theory that considers abrasive and adhesive wear. The model demonstrated improved wear computation accuracy of a contact mechanical seal end face. The validity of the model was established by comparing the simulation results with experimental data and the conventional MB and Archard models. The loading effect and surface morphology parameters involved in wear were also investigated. Results show that severe wear occurs with increasing load ratio, large fractal dimensions, and a higher scale coefficient of the surface morphology. The findings offer a novel method for precisely calculating and projecting the amount of wear of a contact mechanical seal and predicting its wear behavior.

Improved traffic sign recognition algorithm based on YOLOv4-tiny

This study offers an enhanced yolov4-tiny traffic sign identification method for easy deployment on mobile or embedded devices to address the difficulties of a high number of parameters, low recognition accuracy, and poor real-time performance of traffic sign recognition models in complex scenarios. The yolov4-tiny network serves as the model’s foundation. To begin, Octave Convolution is incorporated into the backbone network to eliminate low-frequency feature redundancy, lowering the number of parameters in the model and enhancing computational efficiency. Second, the convolutional block attention module is employed to improve the recognition accuracy of small and medium-sized targets by strengthening the weights of traffic sign regions and suppressing the weights of invalid features. Finally, in the feature fusion stage, the Feature Pyramid Networks structure is replaced with the Simplified Path Aggregation Network structure to improve the fusing of shallow feature information with deep semantic knowledge and lower the miss detection rate even more On the TT100K data set as well as on CCTSDB dataset, the experimental results suggest that our technique can achieve good recognition performance. With a 16MB model size, our solution improves the mean average precision by 3.5 percent and the Frame Per Second by 12.5 f/s when compared to the yolov4-tiny algorithm. Our method outperforms yolov4-tiny in terms of recognition accuracy and detection speed, and it can easily meet the real-time requirements for traffic sign recognition.

Improved YOLOv7-based algorithm for elevator passenger detection

The task of identifying the number of passengers in an elevator plays an important role in optimizing the elevator algorithm and ensuring the safety of elevator passengers. In order to enable the elevator passenger-carrying algorithm to be deployed in edge devices, a lightweight elevator passenger-carrying algorithm based on improved yolov7 is designed, which reduces the model size while ensuring high precision. The proposed algorithm adopts a more lightweight attention mechanism structure in the head of the network. And made the lift-person-detection dataset as the experimental dataset. The dataset has nearly 2,000 image samples, including 900 training set images and 600 test set images. Experiments show that the proposed model achieves 98.9% mAP recognition accuracy. Compared with the 71.21MB model size of the original yolov7 network, the model size of the model proposed in this paper is reduced to 63.24MB, and the volume is reduced by 11.13%.

Algorithm for Determining the Purity of Pure Organic Phthalates by the Indirect Mass Balance Method

Pure organic substances are the first and key link in the chain of metrological traceability in organic analysis. In most cases, purity determination of the organic substance is a non-trivial task, the solution of which requires preliminary theoretical study and understanding of the capabilities of instrumental methods for developing a pertinent analytical procedure.Within the study, an algorithm for determining the purity of a separate group of organic compounds – ortho-phthalic acid esters (phthalates) was developed based on the mass balance model and the general algorithm for characterizing substances suitable for isolation or purification by distillation. An additional classification of probable impurities (into five types A – E) depending on their qualitative and quantitative characteristics is proposed, as well as suitable analytical methods for measuring impurities and confirming the identity of the main component are identified.The proposed algorithm was implemented on the State Primary Standard GET 208 to determine the purity of six priority phthalates: dimethyl phthalate, diethyl phthalate, di-(n-butyl) phthalate, benzyl butyl phthalate, di-(2-ethylhexyl) phthalate, and di-(n-octyl) phthalate. Certified pure phthalates were used to create a certified reference material for the composition of a solution of orthophthalic acid esters (phthalates) in methanol (6Ftlt-VNIIM) GSO 11366–2019.The study revealed that the adapted algorithm considers the specifics of work with the phthalates group and the features of the separation of probable impurities within the MB model, and is universal for all congeners of the group, but does not exclude the natural constraints of the mass balance method.

Feature distillation Siamese networks for object tracking

In recent years, Siamese-based trackers have shown remarkable improvement in visual tracking. The general trends are interested in making deeper and more complicated networks to pursue higher accuracy. However, these advances result in cumbersome trackers with respect to size and speed, which hinders the deployment of deep trackers on edge devices. Due to the tracking scenario complexity and temporal coherence, the backbone network of deep trackers emphasizes the target appearance information more. However, the appearance feature is sensitive to parameter variation, making the traditional deep model compression methods hard to compress a deep tracker. To bridge the gap between deep Siamese trackers and practical use, we propose a new feature distillation algorithm suitable for deep trackers in this paper. Firstly, motivated by the concept of divide-and-conquer, we formulate the feature distillation into a stepwise distillation problem and perform distillation on each minimum unit to relieve the hard-to-distill problem of appearance feature. Secondly, we reconstruct the student model into a combination of convolution kernels and a point-wise convolutional layer, which enables the student model to inherit all the parameters of the teacher model during initialization. Finally, we propose a 3-step warm-up training strategy to address the student model’s degradation and structural adaptation problems during training. Extensive experiments on eight benchmarks demonstrate that our proposed method compresses the fully convolutional Siamese Networks (SiamFC) and its variant Siamd and achieves leading tracking performance with only 2.1 MB model size while running at 225 fps.

NRP1 inhibition modulates radiosensitivity of medulloblastoma by targeting cancer stem cells

BackgroundMedulloblastoma (MB) is the most common pediatric malignant brain tumor. Despite current therapies, the morbidity and recurrent risk remains significant. Neuropilin-1 receptor (NRP1) has been implicated in the tumor progression of MB. Our recent study showed that NRP1 inhibition stimulated MB stem cells differentiation. Consequently, we hypothesized that targeting NRP1 in medulloblastoma could improve current treatments.MethodsNRP1 inhibition with a novel peptidomimetic agent, MR438, was evaluated with radiotherapy (RT) in MB models (DAOY, D283-Med and D341-Med) in vitro on cancer stem-like cells as well as in vivo on heterotopic and orthotopic xenografts.ResultsWe show that NRP1 inhibition by MR438 radiosensitizes MB stem-like cells in vitro. In heterotopic DAOY models, MR438 improves RT efficacy as measured by tumor growth and mouse survival. In addition, clonogenic assays after tumor dissociation showed a significant reduction in cancer stem cells with the combination treatment. In the same way, a benefit of the combined therapy was observed in the orthotopic model only for a low cumulative irradiation dose of 10 Gy but not for 20 Gy.ConclusionsFinally, our results demonstrated that targeting NRP1 with MR438 could be a potential new strategy and could limit MB progression by decreasing the stem cell number while reducing the radiation dose.

CSIG-31. T-TYPE CALCIUM CHANNEL BLOCKERS INDUCE METABOLIC ALTERATIONS AND APOPTOSIS IN MEDULLOBLASTOMA CELLS

Abstract Medulloblastoma (MB) is the most common malignant paediatric brain tumour. In our previous studies, in a screen of a panel of calcium channel modulators in our novel high-throughput 3D assay, we identified T-type (CaV3) channel inhibitors, mibefradil and NNC-55-0396 (NNC) as selective inhibitors of MB cell growth. Mibefradil was originally approved for the treatment of hypertension and angina pectoris, and recently successfully completed a Phase-I trial for recurrent high-grade glioma. NNC is an analogue of mibefradil with multiple advantages compared to mibefradil that makes it attractive for potential future clinical trials. T-type channels have a unique low voltage-dependent activation/inactivation, and many studies suggest that they have a direct regulatory role in controlling Ca2+ signalling in non-excitable tissues, including cancers. In this study, we characterized the effect of mibefradil and NNC on MB cells and elucidated their mechanism of action. This study demonstrates that the induction of toxicity in MB cells is selective to T-type but not to L-type Ca2+ channel inhibitors. Addition of CaV3 inhibitors to vincristine sensitised MB cells to this MB chemotherapeutic agent, suggesting an additive effect. Furthermore, CaV3 inhibitors induced cell death in MB cells via apoptosis. Supported by proteomics data and cellular assays, apoptotic cell death was associated with reduced mitochondrial membrane potential and reduced ATP levels, which suggests that both compounds alter the metabolism of MB cells. We also identified that CACNA1H (encoding the CaV3.2 protein) is significantly upregulated in aggressive Groups 3 and 4 MB patient samples, and its higher expression is associated with higher metastasis and lower patient survival rates. This study offers new insights into the action of mibefradil and NNC and will pave the way to test these molecules or their analogues in pre-clinical MB models alone and in combination with vincristine to assess their suitability as a potential MB therapy.