NTU Journal of Engineering and Technology

Liver disease is one of the major health threats throughout the world. Among the different imaging modalities involved in diagnosing and managing liver conditions, Magnetic resonance imaging (MRI) and computerized tomography (CT) scan play a pivotal role. Various studies were focused on developing automated systems to detect and classify liver diseases using advanced image processing and machine learning algorithms. A review of the literature shows that machine learning models are capable of predicting liver disorders from MRI and CT images. Research indicates that deep learning techniques, especially convolutional neural networks (CNNs), surpass traditional approaches in the extraction and classification of features, thereby enhancing diagnostic accuracy and facilitating early disease detection. This study advances global healthcare initiatives by employing machine learning to enhance the precision of liver disease diagnosis and treatment, aligning with (Goal 3: Good Health and Well-Being). This also advances medical technology by fostering innovation in medical imaging and incorporating AI-driven solutions into healthcare systems (Goal 9: Industry, Innovation, and Infrastructure).

Detection of muscle fatigue is present in many fields, from the training of athletes and the aid in rehabilitation to the control of prosthetics. It can be considered ongoing research aiming to understand the human body and aid in protecting it. Fatigue itself can’t be measured directly but can be detected through other means, for example, electromyography, more specifically surface electromyography (sEMG), which is considered a noninvasive and easy-to-use tool. There are various studies on sEMG and its use in detecting muscle fatigue. This study presents a closer look at the application of sEMG in muscle fatigue detection, as well as the specifications required when dealing with it, such as muscle signal preprocessing, analysis methods, and post-processing, including feature extraction. This research serves as a simple guide for individuals that are new to this topic, enabling them to have a comprehensive view about it.

The persistent challenge of achieving accurate, painless, and continuous blood glucose monitoring in diabetes management has driven intensive research toward non-invasive technological solutions that balance precision, accessibility, and patient comfort. This work examines recent advancements in sensor-based glucose estimation and emphasizes the strong synergy between modern sensing technologies and deep learning methodologies. Emerging techniques such as bioimpedance analysis, near-infrared (NIR), and mid-infrared (Mid-IR) spectroscopy have demonstrated promising potential for glucose monitoring without the need for invasive blood sampling. In parallel, hybrid sensor architectures combined with artificial intelligence–driven predictive models have shown the ability to mitigate traditional limitations and enhance real-time monitoring performance. Despite these advances, significant challenges remain, including environmental sensitivity, inter-individual physiological variability, and calibration stability. To address these issues, adaptive machine learning frameworks are being developed to improve system robustness and reliability. This review critically evaluates current progress, identifies existing limitations, and outlines future research directions toward clinically viable, FDA-compliant non-invasive glucose monitoring systems.

The advancements in assistive audio recognition technologies have progressed rapidly and have improved the accessibility for the Deaf and Hard-of-Hearing (DHH) community. This paper provides an extensive review of the current methods, including traditional techniques such as GMMs and HMMs, and modern deep learning-based techniques as Convolutional Neural Networks (CNN) and EfficientNet. The strengths and limitations of these methods are compared with respect to accuracy, computational efficiency, and noise robustness. Although these deep learning models have higher recognition rates, they impose significant computational requirements thereby restricting their usage in real-time and low-power devices. The review also revealed research gaps with a strong emphasis on energy-aware neural networks and their potential applications in smart environments, adaption strategies, and integration of IoT technol-ogies for practical implementation. As deep learning models grow in complexity and the required amount of labeled data increased to unsustainable level, future research will need to explore hybrid models that balance between performance gains and efficiency, to ensure that assistive audio recognition systems become increasingly reliable, usable, and generalizable in a broad range of acoustic environments.

Hand gesture detection is essential for improving human-computer interaction, considerably advancing the creation of more intuitive and effective interfaces. This study examines the efficacy of three sophisticated object identification models in identifying hand motions: YOLOv8, YOLOv10, and YOLOv11. A dataset was created for this purpose, enhanced via Roboflow, and utilized for training and evaluating the performance of these models based on metrics including mAP, Precision, Recall, and F1-score. Performance was enhanced by adjusting the optimizers (Adam, SGD, and AdamW) and hyperparameters, including learning rate and epochs. The findings indicate outstanding performance, with YOLOv8 and YOLOv10 attaining mean Average Precisions of 99.2% and 99.1%, respectively, alongside precisions of 98.1% and 98.6%, recalls of 97.7% and 97.2%, YOLOv11 had a mean Average Precision of 99.2%, a precision of 98.6%, and a recall of 98.4%, These findings demonstrate that YOLOv11 adeptly manages more complex datasets while preserving an ideal equilibrium between speed and precision.

A wide variety of medical diagnoses depend on the study and testing of the dis-ease images. Anemia is one of the cases separated in the populations. Deep Learning (DL) is one of the subfields of artificial intelligence (AI) techniques applied in the healthcare system to diagnose diseases. In this work, YOLOv11 with its versions Nano (n), Small (s), Medium (m), Large (l), and Extra-large (x) produces deep-learning models to detect anemia based on the palm images. The models train on 4260 color images of children under 5 years labeled anemic and nonanemic. The models train using two different dataset splits, with two input image sizes, (64x64) and (128x128). The high training accuracy achieves at YOLOv11n 99.3% at the group two dataset with an input image size of 128 and YOLOv11n 98.9% at the group one dataset with an input image size of 128. All models are test, and the best test results obtains with the Yolov11n models, 98.9% and 99.5% of the two groups at input image size 128, which predicted correctly with a high percentage in a very short time. We produce these models to assess medical images, providing precise automated estimations and reducing diagnosis time and errors. Additionally, they help reduce death rates and promote early intervention to enhance the quality of life for patients.

Portable electronic devices must have effective thermal management to maintain low enough working temperatures for dependability, expanded installed functionality, and user comfort. In certain situations, using heat sinks made of metal foam has the potential for heat dissipation. In the research, three cases were employed with and without copper foam, forced air convection into the device's duct, and electric heating of the copper plate under the heat sink. For each of the three heat sink scenarios, which happen again with different porosities throughout the heating and cooling processes, the effects of the thermal power and input air velocity parameters of metal foam were investigated. The findings indicate that the copper plate in case 3, which features two copper foams, has the lowest temperature among the three cases. The temperature dropped during the heating process in cases 3 (two pieces of copper foam) and 2 (one piece of copper foam) at porosity 90% (198%) and 95% porosity (197.6%) for case 3 and 45.2% and 46.8% for case 2, respectively, compared to case 1 (no copper foam). Because of its high heat conductivity, the copper plate in case 3 maintained its temperature during the cooling process. Compared to case 1, the improvement rates in the copper foam example were 86% and 77% lower, respectively. Temperature was less affected by increasing the input air velocity.

The proposed method applies YOLOv10 deep learning to detect license plates with subsequent text extraction from plates through EasyOCR and OCR engines. This system architecture operates in real-time allowing its deployment in unmanned traffic surveillance systems law enforcement detection and automated parking operations. The YOLOv10 model achieves quick and reliable license plate detection through CSPNet and SCDD while working in dynamic settings. The system combines various OCR methods into its architecture for processing text with different font types and challenging visual factors. The model exhibits strong precision-recall abilities and generalization power based on Precision-Recall Curve and F1-Score and Confusion Matrix results which lead to a 0.986 mean Average Precision (mAP). The system functions independently and accepts real-time detection through either static images, moving video, or live source vide. This solution expands to accommodate usage in extensive traffic management systems which enable practical vehicle tracking alongside toll collection and security surveillance operations.

It is common knowledge that hematology test panels include measurements of the packed cell volume (PCV), also known as the hematocrit, because they are so helpful in any hematologic workup. The packed red blood cells volume (PCV) value can be calculated in a variety of ways. Some of them use direct approaches, while others use indirect ones. The most straightforward and cost-effective method, which is frequently used as a direct means to assess PCV but has a chance of human mistake and trapped plasma, is using a manual ruler to evaluate PCV. In our search, we developed and built an electronic circuit to assess PCV that relies on the Arduino board. The novel measurement tool is made up of red laser (650 nm) emitters to take advantage of the red light's specific absorption by erythrocytes (red blood cells), which is higher than the red light's lowest absorption by blood plasma. The photocell then receives the transmitted light, and the Arduino board processes the photocell's signal. The PCV evaluation method is reliable because of the change in light intensity between the incident and transmitted light after passing through a centrifuged blood sample. The device's six lasers and photocells make it easier and more accurate to calculate the ratio of erythrocyte volume to total blood volume. Last but not least, a PCV blood analysis can be made utilizing a programmable Arduino board 2560 mega (with determined mathematical formulae that are stored in Arduino memory to calculate and screen Values).

Face Emotion Recognition (FER) is essential for improving human-computer interaction. Although deep learning has greatly enhanced the accuracy of FER, deploying these models on embedded devices is challenging due to limitations in computational power and memory. This study proposes a lite and effective FER system for real-time implementation in resource-constrained environments. Lite ResNet-50 and MobileNetV2 were compared for the classification of neutral and angry emotions using a carefully selected dataset. The focus was on measuring accuracy, inference speed, and resource efficiency. Real-time testing is also con-ducted to confirm their applicability in practice. The results show that Lite Res-Net-50 outperforms MobileNetV2 in all key areas, achieving an accuracy of 99.8, inference speeds of 229 ms and with 2 FPS. These findings establish Lite Res-Net-50 as the optimal choice for FER on embedded devices, bridging the gap be-tween deep learning advancements and real-world deployment to improve hu-man-computer interaction.

The continually increasing number in individuals experiencing visual impairments globally, coupled with the challenges encountered by the blind in their daily existence, has prompted the advancement of conventional canes and the implementation of sophisticated assistive technologies to improve the autonomy of the blind and facilitate safer mobility sophisticated assistive technologies to improve the Independence of the blind and facilitate safer mobility. This review present recently utilized the incorporation of modern sensor technologies, namely LiDAR (Light Detection and Ranging) and ultrasonic sensors, into smart canes to improve navigation functionalities. LiDAR enables high-precision obstacle identification and environmental mapping, whilst ultrasonic sensors provide dependable operation in bad weather situations. Utilization Internet of Things (IoT) connection to provide real-time data processing, remote support, and GPS tracking. This paper evaluates current developments in smart cane technology, assessing the ad-vantages and drawbacks of different sensor combinations. systems, and wireless communication modules (Wi-Fi, Bluetooth, GSM) in enhancing user experience. The research underscores the difficulties encountered in sensor fusion, energy consumption, and environmental adaptation. Future advancements must prioritize AI integration for improved obstacle detection, energy-efficient designs, and economical implementations to boost the accessibility of assistive technology. This study advances innovation in assistive mobility solutions, enhancing freedom and safety for visually impaired persons.

Nowadays, skin cancer is regarded as one of the deadliest cancers that people can get. Skin cancer can be either benign or malignant, with melanoma being the most surprising of the two types. They include basal and melanoma. The topic of Sustainable Development Goal (SDG) 3: "Good Health and Well-Being" is relevant to this work. In particular, the goal of this manuscript is to lower the early mortality rate from non-communicable diseases (NCDs), such as cancer, by one-third through treatment and prevention. Treatment for melanoma cancer may benefit from early detection. Numerous systems in use today attest to the significant role that computer vision may play in medical image diagnosis. In this work, an offer methodology for classifying medical photos into benign or malignant tumors using machine learning is proposed. First, we organize and process the image data, and then we train and evaluate the model. The first step of the process is to collect image data from pre-made folders that have been separated into training and testing sets, as well as benign and malignant categories. File paths are combined into data frames for each category and set, which are subsequently labeled appropriately. Active Learning V3 + approach was employed at the inception.

This research includes a practical investigation on the impacts of adding phase change materials (PCMs) into metal foam made from copper in heat sinks, as well as using metal foam alone during heating and cooling. This study used RT55 paraffin as a phase change material. Five cases were used in the experiments. Each one contained a varied quantity of PCM and copper metal foam, in addition to the centrifugal fan-forced airflow inside the device duct and electrical heating of a plate made of copper under the thermal dissipator. This study looked at the impact of thermal energy and intake air velocity coefficients on five heat sink cases Throughout the process of heating and process of cooling. The findings indicate that in the situation of copper foam (case 3), The copper plate has a lower temperature relative to the other instances. During the heating procedure, cases 5 (Paraffin partly filled copper foam) and 3 (2-piece foams made of copper) showed percentage temperature reductions of 57.46% and 66.4% respectively, when compared to case 1 (Absent paraffin or copper foam). During the process of cooling, the high conductivity for copper foam results in case 3 having temperatures that are 43.5% lower than that of case 5, which is 34.7%. furthermore, raising the input air velocity includes less of an impact on heating reduction in scenarios involving phase change materials (PCMs).

To examine key areas and emerging trends in Machine learning (ML) research within the field of lung cancer. Methods Research literature on ML in the field of lung cancer was gathered from the Science Citation Index Expanded within the Scopus database. This data was then analyzed for insights on publication years, countries/regions, affiliation, funding sponsor, citations, and keywords. Co-occurrence network graphs were created using the VOSviewer library and CiteSpace tools on the online analysis platform. Result A total of 3,341 pertinent research articles from various nations were selected. From 2023 to 2024, there was a notable rise in the quantity of indexed documents, reaching a zenith of roughly 2000 in 2024. Nevertheless, this expansion was not maintained, and in 2025, the total declined precipitously to around 500. Geographically, India and China dominate research output, with India marginally in the lead. The United States occupies the third position but trails considerably behind the top two, with the United Kingdom in fourth place. Other nations, such as Italy, Canada, South Korea, Japan, Saudi Arabia, and Germany, provide relatively fewer documentation. The findings demonstrate significant research effort in Asia, especially in India and China, while North America and Europe also contribute considerably. Disparities in research output can be ascribed to elements such as funding, institutional backing, and national policies that affect publication trends, underscoring the necessity for additional exploration into the political and economic aspects of academic publishing. This research provides a comprehensive analysis of the challenges and prospective applications of machine learning in digital clinical diagnostic systems (DCDS) for the detection of LC. Conclusions The study identified key focus areas and emerging trends in ML for lung cancer diagnosis and indicating that this technology could greatly improve early detection.

The precise classification and localization of faults in power transmission lines are very important to ensure grid stability and to minimize service interruptions. Traditional methods of fault detection frequently have difficulty with complicated fault scenarios, faults of high impedance, and varying conditions of operation. Throughout this paper, an approach of a new machine learning-based for fault classification and localization in power transmission lines adopting Long Short-Term Memory (LSTM) networks and vibration data analysis is proposed. In contrast with conventional impedance-based and methods of traveling wave, the proposed approach takes advantage of temporal dependencies within vibration signals for enhancing predictive precision. Through capturing different faults types, simulation as well as location, a dataset of fault conditions is being generated. For recognizing distinguishing fault patterns and predicting location’s fault with high accuracy, LSTM-based model we proposed is trained. The experimental results show how the LSTM model is superior in dealing with sequential data and, subsequently, improves fault localization precision. Moreover, experimental outcomes confirm that the proposed approach is robust, which leads to high classification precision and minimum localization error. The findings show the potentiality of vibration-based machine learning models to revolutionize fault management within power grids, which offers a solution that is more adaptive and data-driven to the challenges of fault classification and localization.

Alzheimer's Disease (AD) is considered one of the most prevalent neurological disorders, primarily affecting elderly people and adversely impacting their brain functions. This disease is characterized by the gradual deterioration of cognitive functions, particularly memory, leading to varying information loss levels, typically classified within the framework of dementia. In this context, Artificial Intelligence (AI) methods, especially Convolutional Neural Networks (CNN) and Optuna optimization, have emerged as effective technique to provide efficient approach (or model) for classifying AD levels. The proposed approach is called the Optuna-CNN (O-CNN). It has the ability to identify the optimal CNN architecture by employing the optimization of Optuna. It is designed to classify AD into four levels (or stages): Mild Dementia (MD), Moderate Dementia (MoD), No Dementia (ND) and Very Mild Dementia (VMD). Two datasets are utilized here: the Best Alzheimer's MRI Dataset (BA-MRI) and the Alzheimer's MRI Dataset (AD-MRI), both are obtained from Kaggle platform. Following extensive training and implementations of the O-CNN method, high classification accuracies 94.79% for the AD-MRI dataset and 99.18% for the BA-MRI dataset are achieved.

To examine key areas and emerging directions in artificial Intelligence (AI) research correlated to Antibiotic resistance. Research literature on AI in the field of Antibiotic resistance was gathered from the Science Citation Index expanded, which is part of the Scopes. This data was analyzed to gain insights into publication years, countries/regions, institutions, citations, and keywords. Co-occurrence network graphs were created using the VOSviewer Library, Publish or Perish, and CiteSpace tools on the online analysis platform. The analysis of AI-related antibiotic sensitivity research (2020–2025) highlights China's guiding contribution, fluctuating magazine trends, disparities in institutional output, and a strong research focus on bacterial resistance, biofilms, and quorum sensing, with bibliometric analysis revealing the dominance of antibiotic resistance as a key topic and citation metrics showing rapidly evolving scientific discussions in microbiology and antimicrobial research. The study identified key focus areas and emerging trends in artificial intelligence for antibiotic resistance and the role of certain bacteria on the patient, suggesting that artificial intelligence will play an increasingly prominent role in antibiotic resistance moving forward.

Surface muscle fatigue (MF) is an important area to study especially in medicine and sport. One way to detect and process muscle fatigue is by the use of surface electromyography (sEMG). This study presents a real-time sEMG signal acquisition and processing system designed to detect muscle fatigue during. The study included four participants that performed isometric contractions until muscle fatigue was reached, during which sEMG signals were continually monitored. The acquired sEMG data underwent systematic processing, including filtering, rectification, and feature extraction. Four features were extracted: Root Mean Square (RMS), Mean Absolute Value (MAV), Mean Frequency (MNF), and Median Frequency (MDF). The results show that MNF is the clearest indicator of fatigue in the suggested system. Moreover, RMS and MAV can be helpful in indicating the early signs of fatigue. The selected method is useful for real-time muscle fatigue monitoring without the need for complex algorithms. These results offer a basis for the next research focusing on enhancing real-time sEMG signal processing techniques using industrial sensors.

Cement production is a significant source of CO₂ emissions; therefore, partial replacement of ordinary Portland cement (OPC) with supplementary materials can reduce environmental impact. This study evaluates ternary blended cement incorporating locally available steel slag, calcined clay, and limestone in Iraq as partial OPC replacements. Replacement levels ranged from 30% to 60%. The effects on standard consistency, setting times, and soundness were investigated. Results showed that water demand increased with higher replacement levels, reaching 48% at 60% substitution. Slag replacement (25–30%) increased initial and final setting times by up to 29% and 45%, respectively. In contrast, calcined clay replacement (up to 35%) reduced initial and final setting times by up to 26% and 15%, respectively. Soundness values remained within acceptable limits for all mixes. The findings indicate that high-volume ternary blends can reduce clinker content while maintaining acceptable physical properties, offering environmentally and economically sustainable alternatives to conventional OPC.

The research examined how sisal natural fibers combined with steel fibers with hooked ends influence the mechanical and physical properties of lightweight concrete. The lightweight quality of pumice made it a suitable alternative to regular aggregates, which supported environmental initiatives through insulation applications and partition panels. 70% of natural sand grains were substituted with fine and coarse pumice stock. The testing process separated the pumice into two fractions: the coarse elements stayed on the 12.5 mm screen while the fine pumice material fit through a 4.75 mm screen. Tensile and compressive strength evaluations were performed as well as flexural strength, modulus of elasticity, and dry density testing. The experimental samples included various volume combinations of sisal fibers at 0.25%, 0.5%, and 0.75% together with steel fibers at 1%, 1.25%, and 1.5% levels. The combination of sisal at 0.5% content delivered elevated tensile and flexural strength results, and performance declined due to increased porosity levels. 1.5% steel fibers exhibited the best performance in terms of compressive and flexural strength, which demonstrates their powerful reinforcement properties. Compressive strength decreased because sisal fibers did not bond well with the cement matrix and created additional holes inside the material. When 0.5% sisal particles were added to concrete instead of the standard mix, the tensile strength went up by 5.1% and the flexural strength went up by 7.8%. Concrete reinforced with 1.5% steel fibers showed strength gains of 10.5% and 26.9%, with minor decreases in workability and density. The experimental outcomes identified the best fiber-to-concrete ratio, offering an ideal combination of strength improvement and environmentally friendly practices in lightweight concrete.