Target Features Research Articles

Diverse neuronal activity patterns contribute to the control of distraction in the prefrontal and parietal cortex.

Goal-directed behavior requires the effective suppression of distractions to focus on the task at hand. Although experimental evidence suggests that brain areas in the prefrontal and parietal lobe contribute to the selection of task-relevant and the suppression of task-irrelevant stimuli, how conspicuous distractors are encoded and effectively ignored remains poorly understood. We recorded neuronal responses from 2 regions in the prefrontal and parietal cortex of macaques, the frontal eye fields (FEFs) and the lateral intraparietal (LIP) area, during a visual search task, in the presence and absence of a salient distractor. We found that in both areas, salient distractors are encoded by both response enhancement and suppression by distinct neuronal populations. In FEF, a larger proportion of units displayed suppression of responses to the salient distractor compared to LIP, with suppression effects in FEF being correlated with search time. Moreover, in FEF but not in LIP, the suppression for the salient distractor compared to non-salient distractors that shared the target color could not be accounted for by an enhancement of target features. These results reveal a distinct contribution of FEF in the suppression of salient distractors. Critically, we found that in both areas, the population level representations of the target and singleton locations were not orthogonal, suggesting a mechanism of interference from salient stimuli.

Self-Training Can Reduce Detection False Alarm Rate of High-Resolution Imaging Sonar

Imaging sonar is a primary means of underwater detection, but it faces challenges of high false alarm rates in sonar image target detection due to factors such as reverberation, noise, and resolution. This paper proposes a method to improve the false alarm rate by self-training a deep learning detector on sonar images. Self-training automatically generates proxy classification tasks based on the sonar image target detection dataset, and pre-trains the deep learning detector through these proxy classification tasks to enhance its learning effectiveness of target and background features. This, in turn, improves the detector’s ability to distinguish between targets and backgrounds, thereby reducing the false alarm rate. For the first time, this paper conducts target detection experiments based on deep learning using high-resolution synthetic aperture sonar images at two frequencies. The results show that, under the conditions of equal or higher recall rates, this method can reduce the false alarm rate by 3.91% and 18.50% on 240 kHz and 450 kHz sonar images, respectively, compared to traditional transfer learning methods.

Hybrid contrastive representations and SBO-based samples generation for rotating machinery anomaly detection based on driven-end current signal

Abstract While deep learning methods based on cross-entropy function have made great advancements in maintaining equipment reliability, their learning paradigm appears less suitable for anomaly detection tasks, which focus on extracting specific target features and ignore the discriminability between classes. Besides, incomplete dataset still hinders the robustness of intelligent detection models, especially when the training dataset only contains normal samples. Therefore, a hybrid contrastive representation and soft Brownian offset (SBO)-based samples generation (HCRS) method based on driven-end current signals is proposed to distinguish normal and abnormal samples in incomplete data scenarios where only normal signals are available. In the proposed HCRS detection framework, an autoencoder is initially trained using only normal current samples. It is then combined with the SBO method to generate abnormal samples, relieving the issues of incomplete dataset. Subsequently, a supervised contrastive learning-based deep feature extractor is trained using both the generated abnormal samples and the collected normal samples. This process aims to learn high-level fine-grained representations with discriminability. Finally, these learned representations are utilized to train a data-driven classifier, enabling effective anomaly detection in rotating machinery. In addition, experiments on two datasets suggest that the proposed HCRS can effectively achieve higher accuracy anomaly detection with only normal current signals and outperform existing detection approaches.

Study on Scattering Characteristics of Buildings Under Trees Based on Polarimetric SAR

Accurate detection of buildings is of great significance for urban planning and rapid development. Using the rich scattering information contained in polarimetric SAR, relevant information of target features can be extracted. For buildings covered by vegetation, the penetration capability of long-wavelength SAR can be utilized to further extract their scattering characteristic information, enabling more effective detection. This study analyzes experimental data based on ALOS PALSAR imagery of Shenzhen. The results indicate that buildings obscured by trees exhibit different scattering characteristics compared to other buildings, and analyzing their scattering component information holds great potential for land classification and building detection.

A multi-scale rotated ship targets detection network for remote sensing images in complex scenarios

Detecting ship targets in remote sensing images within complex scenarios faces numerous challenges. The limited feature information of small-scale targets and their random orientation angles often result in missed and false detections. To address these issues, this paper proposes a Multi-Scale Rotated Detection Network (MSRO-Net) for detecting rotated ship targets in remote sensing images. The network adopts a CNN-Transformer hybrid architecture for collaborative feature extraction and integrates our proposed Coordinate-Aware Pyramid Feature Aggregation module (CAPP). This backbone network retains the capability of local feature extraction while also connecting global context and capturing long-range dependencies. The model can gather information from any position in the sequence, extract contextual information of targets at different scales, and enhance global feature representation. Additionally, this paper proposes an Upsampling Feature Reconstruction Pyramid (ARFPN-C), based on Adaptive Rotated Convolution (ARC). This network combines ARC adaptive rotating convolution with a multi-scale feature fusion mechanism to enhance the model’s perceptual capability, addressing the issues of limited target feature information and random orientation angles. The proposed algorithm is validated on two public remote sensing image datasets, HRSC2016 and DOTA. The mAP07, mAP12 on the HRSC2016 dataset, as well as the mAP in the ship category of the DOTA dataset, show a significant advantage over other commonly used object detection algorithms, with accuracies of 90.70%, 98.98%, and 89.46%, respectively. These results further validate the accuracy and effectiveness of the proposed MSRO-Net in object detection.

Improved printed circuit board defect detection scheme

In this paper, an improved printed circuit board(PCB)defect detection scheme named PD-YOLOv8 is proposed, which is specialized in the common and challenging problem of small target recognition in PCB inspection. This improved scheme mainly relies on the basic framework of YOLOv8n, and effectively enhances the detection performance of PCB small defects through multiple innovative designs. First, we incorporate the Efficient Channel Attention Network (ECANet) attention mechanism into the backbone network of YOLOv8, which improves the performance of small-target detection by adaptively enhancing the expressiveness of key features, so that the network possesses higher sensitivity and focus on tiny details in PCB images. Second, we optimize and upgraded the neck structure. On the one hand, the module is introduced to facilitate cross-layer feature fusion to ensure that the rich texture information at the lower layer and the abstract semantic information at the higher layer complement each other, which is conducive to improving the contextual understanding of small target detection; on the other hand, a detection head specialized for small targets is designed and added to enhance the ability of locating and identifying tiny defects. Furthermore, in order to further enhance the interaction and fusion of multi-scale features, we also add a SlimNeck module to the neck structure, which realizes efficient information transfer through streamlined design and reduces computational complexity at the same time. In addition, we draw on the advanced BiFPN structure, which enables the bidirectional flow of feature information between multiple layers and greatly improves the capture and integration of small target features. Compared to the original YOLOv8 algorithm, this algorithm improves the average accuracy on small targets by for mAP50.

Reducing bias in source-free unsupervised domain adaptation for regression.

Due to data privacy and storage concerns, Source-Free Unsupervised Domain Adaptation (SFUDA) focuses on improving an unlabelled target domain by leveraging a pre-trained source model without access to source data. While existing studies attempt to train target models by mitigating biases induced by noisy pseudo labels, they often lack theoretical guarantees for fully reducing biases and have predominantly addressed classification tasks rather than regression ones. To address these gaps, our analysis delves into the generalisation error bound of the target model, aiming to understand the intrinsic limitations of pseudo-label-based SFUDA methods. Theoretical results reveal that biases influencing generalisation error extend beyond the commonly highlighted label inconsistency bias, which denotes the mismatch between pseudo labels and ground truths, and the feature-label mapping bias, which represents the difference between the proxy target regressor and the real target regressor. Equally significant is the feature misalignment bias, indicating the misalignment between the estimated and real target feature distributions. This factor is frequently neglected or not explicitly addressed in current studies. Additionally, the label inconsistency bias can be unbounded in regression due to the continuous label space, further complicating SFUDA for regression tasks. Guided by these theoretical insights, we propose a Bias-Reduced Regression (BRR) method for SFUDA in regression. This method incorporates Feature Distribution Alignment (FDA) to reduce the feature misalignment bias, Hybrid Reliability Evaluation (HRE) to reduce the feature-label mapping bias and pseudo label updating to mitigate the label inconsistency bias. Experiments demonstrate the superior performance of the proposed BRR, and the effectiveness of FDA and HRE in reducing biases for regression tasks in SFUDA.

An image dataset for analyzing tea picking behavior in tea plantations

Tea is an important economic product in China, and tea picking is a key agricultural activity. As the practice of tea picking in China gradually shifts towards intelligent and mechanized methods, artificial intelligence recognition technology has become a crucial tool, showing great potential in recognizing large-scale tea picking operations and various picking behaviors. Constructing a comprehensive database is essential for these advancements. The newly developed Tea Garden Harvest Dataset offers several advantages that have a positive impact on tea garden management: 1) Enhanced image diversity: through advanced data augmentation techniques such as rotation, cropping, enhancement, and flipping, our dataset provides a rich variety of images. This diversity improves the model’s ability to accurately recognize tea picking behaviors under different environments and conditions. 2) Precise annotations: every image in our dataset is meticulously annotated with boundary box coordinates, object categories, and sizes. This detailed annotation helps to better understand the target features, enhancing the model’s learning process and overall performance. 3) Multi-Scale training capability: our dataset supports multi-scale training, allowing the model to adapt to targets of different sizes. This capability ensures versatility and accuracy in real-world applications, where objects may appear at varying distances and scales. This tea garden picking dataset not only fills the existing gap in the data related to tea picking in China but also makes a significant contribution to advancing intelligent tea picking practices. By leveraging its unique advantages, this dataset becomes a powerful resource for tea garden management, promoting increased efficiency, accuracy, and productivity in tea production.

MCF-DTI: Multi-Scale Convolutional Local-Global Feature Fusion for Drug-Target Interaction Prediction.

Predicting drug-target interactions (DTIs) is a crucial step in the development of new drugs and drug repurposing. In this paper, we propose a novel drug-target prediction model called MCF-DTI. The model utilizes the SMILES representation of drugs and the sequence features of targets, employing a multi-scale convolutional neural network (MSCNN) with parallel shared-weight modules to extract features from the drug side. For the target side, it combines MSCNN with Transformer modules to capture both local and global features effectively. The extracted features are then weighted and fused, enabling comprehensive feature representation to enhance the predictive power of the model. Experimental results on the Davis dataset demonstrate that MCF-DTI achieves an AUC of 0.9746 and an AUPR of 0.9542, outperforming other state-of-the-art models. Our case study demonstrates that our model effectively validated several known drug-target relationships in lung cancer and predicted the therapeutic potential of certain preclinical compounds in treating lung cancer. These findings contribute valuable insights for subsequent drug repurposing efforts and novel drug development.

HiRegEx: Interactive Visual Query and Exploration of Multivariate Hierarchical Data.

When using exploratory visual analysis to examine multivariate hierarchical data, users often need to query data to narrow down the scope of analysis. However, formulating effective query expressions remains a challenge for multivariate hierarchical data, particularly when datasets become very large. To address this issue, we develop a declarative grammar, HiRegEx (Hierarchical data Regular Expression), for querying and exploring multivariate hierarchical data. Rooted in the extended multi-level task topology framework for tree visualizations (e-MLTT), HiRegEx delineates three query targets (node, path, and subtree) and two aspects for querying these targets (features and positions), and uses operators developed based on classical regular expressions for query construction. Based on the HiRegEx grammar, we develop an exploratory framework for querying and exploring multivariate hierarchical data and integrate it into the TreeQueryER prototype system. The exploratory framework includes three major components: top-down pattern specification, bottom-up data-driven inquiry, and context-creation data overview. We validate the expressiveness of HiRegEx with the tasks from the e-MLTT framework and showcase the utility and effectiveness of TreeQueryER system through a case study involving expert users in the analysis of a citation tree dataset.

The effects of gender and gender (in)congruency on level-2 visuo-spatial perspective-taking performance: An individual participant data meta-analysis.

Level-2 visuo-spatial perspective-taking (VPT) helps us to understand how the world appears for another person. The process has been linked to conceptual forms of perspective-taking, such as empathic perspective-taking. The present study tested whether similarity to the target of the process, as indicated by gender (in)congruency, affects its embodiment and conclusively answers the question whether there are gender differences in VPT performance. To address these questions, data of N = 2,226 female and male participants, completing K = 107,535 trials of a Level-2 VPT task involving female and male targets, were subjected to an independent participant meta-analysis. Confirmatory analyses revealed that gender (in)congruency did not affect Level-2 VPT performance, speaking against an effect of perceived similarity on the embodiment of Level-2 VPT. Additionally, we observed a significant performance advantage for female participants. Exploratory analyses showed gender-congruency effects can be detected if attentional task demands are low, likely making it easier for participants to process target features such as their gender. These findings clarify the disputed nature of gender differences in Level-2 VPT performance and inform theorizing about embodied and nonembodied strategies used to solve Level-2 VPT tasks, as well as process models of Level-2 VPT performance more generally. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Multi-Source Transfer Learning for zero-shot Structural Damage Detection

Developing generalizable Structural Health Monitoring (SHM) tools for downstream tasks, such as Structural Damage Detection (SDD), is one way to tackle large-scale SHM applications. Such tools become particularly important when dealing with a heterogeneous population of structures lacking prior (damaged and non-damaged) data, a common scenario in large-scale SHM applications. Devising Transfer Learning (TL) mechanisms using representative SHM datasets is a favorable method to address this challenge. Multiple SHM datasets are publicly available but only partially cover the objective representative SHM dataset. The missing parts limit the ability to extract or embed high-level features, resulting in non-generalizable solutions, such as one-to-one (source-to-target) TL models. In the absence of a representative SHM dataset, this study tackles the challenge of accumulating SDD knowledge from sparse, non-similar datasets and integrating them to enable damage detection in unseen target structures with no prior damaged data. The innovation lies in two areas: (1) defining a feature and Domain Adaptation (DA) method that can transfer features between dissimilar structures, and (2) designing an SDD model to learn from these features and a mechanism to combine the learned SDD knowledge from various sources in a zero-shot setting, a necessity in online SHM applications without prior damaged data. To that end, this study proposes a DA-based TL approach and employs ensemble techniques to combine knowledge from parametric models trained on each source dataset, thereby establishing a common ground for fusing data from different structures. The proposed zero-shot Multi-Source TL-SDD technique achieves mean F1 scores of 0.971, 0.959, and 0.922 in three well-known SHM benchmark datasets, without requiring fine-tuning of models pre-trained on source data for the target features. This performance is achieved by integrating signal processing domain knowledge into the DA procedure, outperforming other competitive deep zero-shot SDD techniques trained with the target non-damaged data. Computational complexity analysis is also examined, demonstrating the method satisfying online SHM requirements, a critical feature for real-world applications.

Development and characterization of fluorescent cholesteryl probes with enhanced solvatochromic and pH-sensitive properties for live-cell imaging

We present novel fluorescent cholesteryl probes (CNDs) with a modular design based on the solvatochromic 1,8-phthalimide scaffold. We have explored how different modules—linkers and head groups—affect the ability of these probes to integrate into lipid membranes and how they distribute intracellularly in mouse astrocytes and fibroblasts targeting lysosomes and lipid droplets. Each compound was assessed for its solvatochromic behavior in organic solvents and model membranes. Molecular dynamics simulations and lipid partitioning using giant unilamellar vesicles showed how these analogs behave in model membranes compared to cholesterol. Live-cell imaging demonstrated distinct staining patterns and cellular uptake behaviors, further validating the utility of these probes in biological systems. We compared the empirical results with those of BODIPY-cholesterol, a well-regarded fluorescent cholesterol analog. The internalization efficiency of fluorescent CND probes varies in different cell types and is affected mainly by the head groups. Our results demonstrate that the modular design significantly simplifies the creation of fluorescent cholesteryl probes bearing distinct spectral, biophysical, and cellular targeting features. It is a valuable toolkit for imaging in live cells, measuring cellular membrane dynamics, and studying cholesterol-related processes.

CParty: hierarchically constrained partition function of RNA pseudoknots.

Biologically relevant RNA secondary structures are routinely predicted by efficient dynamic programming algorithms that minimize their free energy. Starting from such algorithms, one can devise partition function algorithms, which enable stochastic perspectives on RNA structure ensembles. As the most prominent example, McCaskill's partition function algorithm is derived from pseudoknot-free energy minimization. While this algorithm became hugely successful for the analysis of pseudoknot-free RNA structure ensembles, as of yet there exists only one pseudoknotted partition function implementation, which covers only simple pseudoknots and comes with a borderline-prohibitive complexity of O(n5) in the RNA length n. Here, we develop a partition function algorithm corresponding to the hierarchical pseudoknot prediction of HFold, which performs exact optimization in a realistic pseudoknot energy model. In consequence, our algorithm CParty carries over HFold's advantages over classical pseudoknot prediction in characterizing the Boltzmann ensemble at equilibrium. Given an RNA sequence S and a pseudoknot-free structure G, CParty computes the partition function over all possibly pseudoknotted density-2 structures G∪G' of S that extend the fixed G by a disjoint pseudoknot-free structure G'. Thus, CParty follows the common hypothesis of hierarchical pseudoknot formation, where pseudoknots form as tertiary contacts only after a first pseudoknot-free "core" G and we call the computed partition function hierarchically constrained (by G). Like HFold, the dynamic programming algorithm CParty is very efficient, achieving the low complexity of the pseudoknot-free algorithm, i.e. cubic time and quadratic space. Finally, by computing pseudoknotted ensemble energies, we unveil kinetics features of a therapeutic target in SARS-CoV-2. CParty is available at https://github.com/HosnaJabbari/CParty.

A Dynamic Measurement System Based on Adaptive Clustering and Multi-Classifier

This technology is highly suitable for detecting and tracking multi-skeletal targets in the field of aerial refueling. The paper presents a target feature responder based on K-means clustering, used for categorizing samples and training models. It employs a decision function to optimize the localization region across various classifier detection and tracking algorithms for targets. Additionally, a state judgment module scores the classifiers based on target state and depth information, which allows for adaptive selection of the classifier. To address the issue of missing target information, the method incorporates a stereo-vision-based mechanism to complete the localization region. This approach effectively handles challenges related to target appearance deformation, significant scale variations, motion blur, and occlusions.

Priming of pop-out in the spatial-cueing paradigm

Abstract Searching for a unique target is faster when its unique feature repeats than when it changes. The standard account for this priming-of-popout (PoP) phenomenon is that selecting a target increases the attentional priority of its features in subsequent searches. However, empirical tests of this priority account have yielded contradictory findings. Here, we attempted to clarify why support of its predictions has been elusive in spatial-cueing studies. We noted that the repetition manipulation is stronger in PoP studies than in spatial-cueing studies: The target and nontarget features either repeat or swap in the former, whereas in the latter, the target color either repeats or changes while the nontarget color remains constant. Here, we implemented the stronger repetition manipulation in the spatial-cueing paradigm. If PoP affects attentional priority, the cue-validity effect should be larger when the current-cue color and previous-target color match than when they mismatch. The results of Experiment 1 revealed the opposite effect. Moreover, they showed that well-replicated contingent-capture effects do not hold when target and nontarget colors are allowed to swap—an observation that was confirmed in Experiment 2. In Experiment 3, we verified that with our experimental set-up, a subtle manipulation of attentional priority could modulate cue-validity effects. While the present study does not resolve why evidence for the priority account is inconsistent with the spatial-cueing paradigm, they eliminate one possible reason for such inconsistency. They also reveal that allowing the target and distractors color to swap masks contingent-capture effects, thereby highlighting the complexity of spatial-cueing effects.

Lightweight Research on Fatigue Driving Face Detection Based on YOLOv8

Introduction: With the rapid development of society, motor vehicles have become one of the main means of transportation. However, as the number of motor vehicles continues to increase, traffic safety accidents also continue to appear, bringing serious threats to people's lives, and property safety. Fatigue driving is one of the important causes of traffic safety accidents. Method: To address this problem, a target detection algorithm called VA-YOLO is designed to improve the speed and accuracy of facial recognition for fatigue checking. The algorithm employs a lightweight backbone network, VanillaNet, instead of the traditional backbone network, which reduces the computational and parametric quantities of the model. The SE attention mechanism is also introduced to enhance the model's attention to the target features, which further improves the accuracy of target detection. Finally, in terms of the bounding box regression loss function, the SIoU loss function is used to reduce the error. Result: The experimental results show that, compared toYolov8n, the VA-YOLO algorithm improves the accuracy by 1.3% while the number of parameters decreases by 30%. Conclusion: This shows that the VA-YOLO algorithm has a significant advantage in realizing the balance between the number of parameters and accuracy, which is important for improving the speed and accuracy of fatigue driving detection.

G-RCenterNet: Reinforced CenterNet for Robotic Arm Grasp Detection.

In industrial applications, robotic arm grasp detection tasks frequently suffer from inadequate accuracy and success rates, which result in reduced operational efficiency. Although existing methods have achieved some success, limitations remain in terms of detection accuracy, real-time performance, and generalization ability. To address these challenges, this paper proposes an enhanced grasp detection model, G-RCenterNet, based on the CenterNet framework. First, a channel and spatial attention mechanism is introduced to improve the network's capability to extract target features, significantly enhancing grasp detection performance in complex backgrounds. Second, an efficient attention module search strategy is proposed to replace traditional fully connected layer structures, which not only increases detection accuracy but also reduces computational overhead. Additionally, the GSConv module is incorporated during the prediction decoding phase to accelerate inference speed while maintaining high accuracy, further improving real-time performance. Finally, ResNet50 is selected as the backbone network, and a custom loss function is designed specifically for grasp detection tasks, which significantly enhances the model's ability to predict feasible grasp boxes. The proposed G-RCenterNet algorithm is embedded into a robotic grasping system, where a structured light depth camera captures target images, and the grasp detection network predicts the optimal grasp box. Experimental results based on the Cornell Grasp Dataset and real-world scenarios demonstrate that the G-RCenterNet model performs robustly in grasp detection tasks, achieving accurate and efficient target grasp detection suitable for practical applications.

Biomimetic astrocyte cell membrane-fused nanovesicles for protecting neurovascular units in hypoxic ischemic encephalopathy

Hypoxic ischemic encephalopathy (HIE) refers to neonatal hypoxic brain injury caused by severe asphyxia during the perinatal period. With a high incidence rate and poor prognosis, HIE accounts for 2.4% of the global disease burden, imposing a heavy burden on families and society. Current clinical treatment for HIE primarily focuses on symptomatic management and supportive care. Therefore, the developments of effective treatment strategies and new drug formulations are critical for improving the prognosis of HIE patients. In order to protect the compromised neurovascular units after HIE, we prepared membrane-fused nanovesicles for delivering rapamycin and si EDN1 (TRCAM@RAPA@si EDN1). Due to the homotypic targeting feature of membrane-fused nanovesicles, we employed astrocyte membranes as synthetic materials to improve the targeting of astrocytes in brain while reducing the clearance of nanovesicles by circulatory system. Additionally, the surface of cell membrane was modified with CXCR3 receptors, enhancing the homing of nanovesicles to infarcted lesions. Lipid vesicles were modified with TK and RVG29 transmembrane peptides, enabling responsive release of internal drugs and blood-brain barrier penetration. Internally loaded rapamycin could promote protective autophagy in astrocytes, improve cellular oxidative stress, while si EDN1 could reduce the expression level of endothelin gene, thereby reducing secondary damage to neurovascular units.Graphical

Exosome-mediated delivery of CRISPR-Cas9: A revolutionary approach to cancer gene editing.

Several molecular strategies based on targeted gene delivery systems have been developed in recent years; however, the CRISPR-Cas9 technology introduced a new era of targeted gene editing, precisely modifying oncogenes, tumor suppressor genes, and other regulatory genes involved in carcinogenesis. However, efficiently and safely delivering CRISPR-Cas9 to cancer cells across the cell membrane and the nucleus is still challenging. Using viral vectors and nanoparticles presents issues of immunogenicity, off-target effects, and low targeting affinity. Naturally, extracellular vesicles called exosomes have garnered the most attention as delivery vehicles in oncology-related CRISPR-Cas9 calls due to their biocompatibility, loading capacity, and inherent targeting features. The following review discusses the current progress in using exosomes to deliver CRISPR-Cas9 components, the approaches to load the CRISPR components into exosomes, and the modification of exosomes to increase stability and tumor-targeted delivery. We discuss the latest strategies in targeting recently accomplished in the exosome field, including modifying the surface of exosomes to enhance their internalization by cancer cells, as well as the measures taken to overcome the impacts of TME on delivery efficiency. Focusing on in vitro and in vivo experimentation, this review shows that exosome-mediated CRISPR-Cas9 can potentially treat cancer types, including pancreatic, lymphoma, and leukemia, for given gene targets. This paper compares exosome-mediated delivery and conventional vectors regarding safety, immune response, and targeting ability. Last but not least, we present the major drawbacks and potential development of the seemingly promising field of exosome engineering in gene editing, with references to CRISPR technologies and applications that may help make the target exosomes therapeutic in oncology.