Kheng Cher Yeo

Most computational studies in medical imaging are conducted by researchers with IT or engineering backgrounds, who may lack the clinical and anatomical expertise of medical experts. As a result, important anatomical details, clinically relevant features, and medical context are often underrepresented. This highlights the need for anatomy-guided literature reviews that integrate clinical insight with technical requirements to support the development of more generalizable and clinically meaningful methods. To address this gap, we propose a novel seven-component anatomy-guided literature review framework that links anatomical and clinical relevance with the technical aspects required to develop robust computational methods (3D) tailored to specific diseases and imaging modalities for the extraction of clinically important anatomical features. We apply the framework to Chronic Respiratory Diseases (CRDs), which cause progressive and irreversible lung damage and are a major contributor to respiratory failure and mortality in adults. Importantly, many adult CRDs are recognized to originate early in life during critical phases of lung growth and development in children. Despite this, most computational studies related to CRDs are dominated by adult cohorts, with limited focus on pediatric populations, particularly Indigenous children. The framework is demonstrated using key clinically relevant features of adult bronchiectasis (airway dilatation) and pediatric bronchiectasis (lower-lobe airway dilatation), which is highly prevalent among indigenous children. The framework integrates 3 core computational methods (airway tree extraction, artery-vein separation and extraction, and lobar and segmental localization), approx. 10 key technical aspects (branch detection, tree-length detection, leakage control, completeness and connectivity, mean-surface distance, etc.), and literature from the past decade (7 sources, approx. 20 keywords, approx. 52 publications reviewed, and 15 selected). This process enabled the identification of key limitations and 30 overall research gaps, anatomically and clinically grounded across the selected CRDs and features. Overall, the proposed framework provides a novel way for developing robust, clinically meaningful computational methods applicable to children and anatomically diverse populations across CRDs and other diseases.

Insect classification is important for agricultural management and ecological research, as it directly affects crop health and production. However, this task remains challenging due to the complex characteristics of insects, class imbalance, and large-scale datasets. To address these issues, we propose BioAutoML-NAS, the first BioAutoML model using multimodal data, including images, and metadata, which applies neural architecture search (NAS) for images to automatically learn the best operations for each connection within each cell. Multiple cells are stacked to form the full network, each extracting detailed image feature representations. A multimodal fusion module combines image embeddings with metadata, allowing the model to use both visual and categorical biological information to classify insects. An alternating bi-level optimization training strategy jointly updates network weights and architecture parameters, while zero operations remove less important connections, producing sparse, efficient, and high-performing architectures. Extensive evaluation on the BIOSCAN-5M dataset demonstrates that BioAutoML-NAS achieves 96.81% accuracy, 97.46% precision, 96.81% recall, and a 97.05% F1 score, outperforming state-of-the-art transfer learning, transformer, AutoML, and NAS methods by approximately 16%, 10%, and 8% respectively. Further validation on the Insects-1M dataset obtains 93.25% accuracy, 93.71% precision, 92.74% recall, and a 93.22% F1 score. These results demonstrate that BioAutoML-NAS provides accurate, confident insect classification that supports modern sustainable farming.

In the agricultural domain, the availability of labelled data for disease recognition tasks is often limited due to the cost and expertise required for annotation. In this paper, a novel semi-supervised learning framework named TOM-SSL is proposed for automatic tomato leaf disease recognition using pseudo-labelling. TOM-SSL effectively addresses the challenge of limited labelled data by leveraging a small labelled subset and confidently pseudo-labelled samples from a large pool of unlabelled data to improve classification performance. Utilising only 10% of the labelled data, the proposed framework with a MobileNetV3-Small backbone achieves the best accuracy at 72.51% on the tomato subset of the PlantVillage dataset and 70.87% on the Taiwan tomato leaf disease dataset across 10 disease categories in PlantVillage and 6 in the Taiwan dataset. While achieving recognition performance on par with current state-of-the-art supervised methods, notably, the proposed approach offers a tenfold enhancement in label efficiency.

This comprehensive systematic review examines the integration of software-defined networking (SDN) with IoT security frameworks, analyzing recent advancements in encryption, authentication, access control techniques, and intrusion detection systems. Our analysis reveals that while SDN demonstrates promising capabilities in enhancing IoT security through centralized control and dynamic policy enforcement, several critical limitations persist, particularly in scalability and real-world validation. As intrusion detection represents an integral security requirement for robust IoT frameworks, we conduct an in-depth evaluation of Machine Learning (ML) and Deep Learning (DL) techniques that have emerged as predominant approaches for threat detection in SDN-enabled IoT environments. The review categorizes and analyzes these ML/DL implementations across various architectural paradigms, identifying patterns in their effectiveness for different security contexts. Furthermore, recognizing that the performance of these ML/DL models critically depends on training data quality, we evaluate existing IoT security datasets, identifying significant gaps in representing contemporary attack vectors and realistic IoT environments. A key finding indicates that hybrid architectures integrating cloud–edge–fog computing demonstrate superior performance in distributing security workloads compared to single-tier implementations. Based on this systematic analysis, we propose key future research directions, including adaptive zero-trust architectures, federated machine learning for distributed security, and comprehensive dataset creation methodologies, that address current limitations in IoT security research.

Bronchiectasis is a chronic respiratory disorder characterized by the dilation and damage of bronchial walls due to recurrent and prolonged episodes of inflammation and infection. The diagnosis of bronchiectasis requires objective measurement of abnormal bronchial dilation. It is confirmed radiologically using a chest tomography (C.T.). scan where the pathognomonic feature is increased broncho-arterial ratio (BAR) (>0.8 in children) and, often accompanied by other features such as bronchial wall thickening (W.T.). Developing image processing-based methods facilitates quicker interpretation of the scans and enable detailed evaluations according to the lobes and segments. However, challenges, such as inclined and oblique orientations of structures, and partial volume effect, can complicate accurate measurements in the upper and middle lobes using the same algorithms. Therefore, the accurate detection and measurement of airway and artery regions for BAR and wall thickness in each lobe require specialized image processing/machine learning methods and approaches. Here, we adopt a step-by-step approach and propose methods for three steps: 1. Separating right lower lobe (RLL) region from full-length C.T. scans using tracheal bifurcation (Carina) point as a central marker; 2. Updated technique to locate inner diameter of airways and outer diameter of arteries for BAR measurement; and 3. Measuring airway wall thickness (W.T.) by identifying the outer and inner diameter of airway boundaries (perimeter). Our analysis of 13 high resolution C.T. scans (HRCT) with varying thicknesses (0.67mm, 1mm, 2mm) demonstrates that the frame containing the tracheal bifurcation can be detected accurately in most cases, with a deviation of ±2 frames in some cases. Similarly, a Windows app is developed for measuring inner airway diameter, artery diameter, BAR, and wall thickness, allowing users to draw bounding around visible discrete B.A. pairs in the RLL region. Measurements of 10 B.A. pairs revealed accurate results comparable to those of a human reader, with deviations of ±0.10-0.15mm observed across all measurements. Additional studies and validation are necessary to consolidate inter- and intra-rater variability and enhance the proposed methods.