Colorectal polyps are precancerous lesions with a high risk of developing into cancer if left untreated. Colonoscopy is the gold standard for detecting and removing these polyps, but it has high miss rates, especially for small and flat polyps. Deep learning methods have been increasingly used to aid in polyp detection, but their performance remains limited when applied to samples captured under unconstrained conditions or when processing images of small and flat polyps. To address this challenge, we propose PolyNeXt, a novel polyp segmentation network that leverages ConvNeXtV2 and global response normalization (GRN) layers. We train PolyNeXt on a combined dataset of samples from Kvasir-SEG and CVC-ClinicDB and evaluate it on four public distinct datasets: ETIS-LaribPolypDB, CVC-ColonDB, CVC-300, and BKAI-IGH NeoPolyp-Small. PolyNeXt achieves state-of-the-art performance on polyps captured in suboptimal conditions, outperforming other methods in terms of Intersection over Union (IoU) and Dice coefficient. Our work demonstrates that PolyNeXt is an effective polyp segmentation network that can improve the accuracy and reliability of detecting polyps captured under challenging circumstances.

Colorectal cancer remains one of the most lethal malignancies, underscoring the importance of automated polyp detection for early diagnosis. Although recent multi-scale deep learning models have demonstrated success by integrating local and global feature extraction, they face notable limitations: (1) noisy feature extraction caused by white light reflections and inconsistent feature scales, (2) insufficiently defined border information leading to higher miss rates for flat and irregularly shaped polyps, and (3) limited availability of labeled training data, which hinders generalization and convergence. To address these challenges, we propose a novel polyp segmentation network incorporating a noise mask-guided feature denoising module and a finite-differences-based border enhancement module to improve feature quality and border representation without excessive computational overhead. Furthermore, we introduce a selfsupervised pretext task that extracts meaningful representations from unlabeled colonoscopy images, enhancing model performance on the downstream polyp segmentation task. Our model, pre-trained on LDPolypVideo and fine-tuned on Kvasir-SEG and CVC-ClinicDB, demonstrates state-of-the-art performance across four public datasets (ETIS-LaribPolypDB and CVC-ColonDB), surpassing existing approaches in Intersection over Union (IoU), Dice Coefficient, and computational efficiency.

ABSTRACT Current colorectal polyps detection methods often struggle with efficiency and boundary precision, especially when dealing with polyps of complex shapes and sizes. Traditional techniques may fail to precisely define the boundaries of these polyps, leading to suboptimal detection rates. Furthermore, flat and small polyps often blend into the background due to their low contrast against the mucosal wall, making them even more challenging to detect. To address these challenges, we introduce SCABNet, a novel deep learning architecture for the efficient detection of colorectal polyps. SCABNet employs an encoder‐decoder structure with three novel blocks: the Feature Enhancement Block (FEB), the Channel Prioritization Block (CPB), and the Spatial‐Gradient Boundary Attention Block (SGBAB). The FEB applies dilation and spatial attention to high‐level features, enhancing their discriminative power and improving the model's ability to capture complex patterns. The CPB, an efficient alternative to traditional channel attention blocks, assigns prioritization weights to diverse feature channels. The SGBAB replaces conventional boundary attention mechanisms with a more efficient solution that focuses on the spatial attention of the feature map. It employs a Jacobian‐based approach to construct learned convolutions on both vertical and horizontal components of the feature map. This allows the SGBAB to effectively understand the changes in the feature map across different spatial locations, which is crucial for detecting the boundaries of complex‐shaped polyps. These blocks are strategically embedded within the network's skip connections, enhancing the model's boundary detection capabilities without imposing excessive computational demands. They exploit and enhance features at three levels: high, mid, and low, thereby ensuring the detection of a wide range of polyps. SCABNet has been trained on the Kvasir‐SEG and CVC‐ClinicDB datasets and evaluated on multiple datasets, demonstrating superior results. The code is available on: https://github.com/KhaledELKarazle97/SCABNet .

This study addresses the need for effective cross-selling in the insurance sector by developing a predictive framework to identify health insurance policyholders likely to purchase vehicle insurance. Leveraging AI techniques, the research aims to provide insurers with tools for targeted marketing and personalized recommendations. A subset of 3,000 records from a public dataset (381,000 samples, 10 features) was analyzed, revealing significant data imbalance. To address this, three data synthesis techniques-SMOTE, SMOTENC, and KMeansSMOTE-were applied. Dimensionality reduction methods (PCA and UMAP) were used for visualization, and feature ranking was conducted using RFC to assess feature importance. By tackling class imbalance and employing advanced visualization and feature analysis, this study enhances the accuracy and fairness of insurance recommendations, benefiting both insurers and policyholders in a data-driven landscape.

This study investigates the adverse effects of class imbalance on classification models within healthcare, recognizing the critical importance of addressing this significant challenge in medical datasets and the essential role of precise diagnostic tools in the field. Class imbalance often occurs due to a lack of data for specific classes, such as rare diseases. Our research aims to determine the efficacy of advanced techniques like data augmentation, DCGAN, Pix2Pix, and diffusion methods in addressing the challenge of class imbalance in medical datasets. Using a subset of the HAM10000K dataset comprising 600 normal skin images and 600 melanoma images, the study gradually reduces the number of melanoma images to demonstrate the detrimental effects of class imbalance on classification accuracy. Additionally, the research provides insights into enhancing classification model performance in medical AI applications and challenges in the experimentation technique. The experiment results showcase promising enhancements in the classification model, particularly notable when employing Pix2Pix and Stable Diffusions. Each method exhibits distinct strengths and considerations, contributing to a comprehensive understanding of their efficacy. Future research endeavors should prioritize surmounting the challenges linked with Stable Diffusion. Moreover, exploring its potential for generating multi-modality medical images could unlock new avenues for enhancing classification model performance in imbalanced datasets.

COVID-19 has become a global pandemic with a high mortality rate, and chest X-ray (CXR) is an essential tool for its diagnosis and severity classification. The use of AI in medical image analysis has shown great potential in improving the accuracy and speed of COVID-19 diagnosis. In this study, we propose a two-stages Vision Transformer (ViT)-based architecture for the three-class severity classification of COVID-19 cases using CXR images. The proposed architecture consists of two ViT models, ViT A and ViT B. ViT A performs initial classification between Mild-Intermediate and Advanced classes, while ViT B further classifies CXR images that are classified as Mild-Intermediate into Mild and Intermediate classes. The proposed architecture has achieved an overall accuracy of 0.7933 on the test set, demonstrating its effectiveness in accurately classifying the severity of COVID-19 cases. Additionally, we evaluated the generalization of our model using a publicly available dataset and obtained an overall accuracy of 0.6919, indicating that our model has a good level of generalization. In order to achieve explainability and transparency, we applied Grad-CAM, Score-CAM, and XGrad-CAM to the ViT used in the architecture to provide insights into the model’s decision-making process and identify areas for improvement. The proposed ViT-based architecture achieved high accuracy and robust explainability, which are essential requirements for the deployment of AI-based systems in clinical settings.

In the field of Computer-Aided Detection (CADx), the use of AI-based algorithms for disease detection in endoscopy images, especially colonoscopy images, is on the rise. However, these algorithms often encounter performance issues due to obstructions like specular reflection, resulting in false positives. This paper presents a novel algorithm specifically designed to tackle the challenges posed by high specular reflection regions in colonoscopy images. The proposed algorithm identifies these regions and applies precise inpainting for restoration. The process entails converting the input image from RGB to HSV color space and focusing on the Saturation (S) component in convex regions detected using a Hessian-based method. This step creates a binary mask that pinpoints areas of specular reflection. The inpainting function then uses this mask to guide the restoration of these identified regions and their borders. To ensure a seamless blend of the restored regions with the background and adjacent pixels, a feathering process is applied to the repaired regions. This enhances both the accuracy and aesthetic coherence of the inpainted images. The performance of our algorithm was rigorously tested on five unique colonoscopy datasets and various endoscopy images from the Kvasir dataset, using an extensive set of evaluation metrics and a comparative analysis with existing methods consistently highlighted the superior performance of our algorithm.

The urgent need for early detection and intervention in colorectal cancer highlights the importance of automating colorectal polyp detection. Despite the use of deep learning techniques for polyp detection from medical images, these methods often struggle with issues such as skewed data distribution and class imbalances. In this study, we introduce PolyLite, a lightweight transformer-based model designed to balance performance and efficiency. PolyLite features a custom SegFormer encoder that takes advantage of its lightweight design, and a decoder that uses a multi-scale attention network (MA-NET) to achieve an optimal balance between accuracy and efficiency. To improve accuracy and reduce false negatives, we incorporate a reverse attention layer that effectively counteracts input mask attention during training. This refocuses the model's efforts to better learn non-polyp regions. PolyLite undergoes a two-stage training process, beginning with pre-training on the ImageNet dataset and then fine-tuning on the Kvasir-SEG dataset. We conduct rigorous evaluations on three complex datasets: CVC-ClinicDB, CVC-ColonDB, and ETIS-LaribPolypDB. Empirical results demonstrate PolyLite's superiority over existing techniques, as evidenced by its performance in terms of Intersection over Union and Dice scores. This validates its effectiveness and scalability in polyp segmentation, making complex and computationally intensive architectures unnecessary.

Colorectal polyps is a prevalent medical condition that could lead to colorectal cancer, a leading cause of cancer-related mortality globally, if left undiagnosed. Colonoscopy remains the gold standard for detection and diagnosis of colorectal neoplasia; however, a significant proportion of neoplastic lesions are missed during routine examinations, particularly diminutive and flat lesions. Deep learning techniques have been employed to improve polyp detection rates in colonoscopy images and have proven successful in reducing the miss rate. However, accurate segmentation of small and flat polyps remains a major challenge to existing models as they struggle to differentiate polypoid and non-polypoid regions apart. To address this issue, we present an enhanced version of the Multi-Scale Attention Network (MA-NET) that incorporates a modified Mix-ViT transformer as the feature extractor. The modified Mix-ViT facilitates ultra-fine-grained visual categorization to improve the segmentation accuracy of polypoid and non-polypoid regions. Additionally, we introduce a pre-processing layer that performs histogram equalization on input images in the CIEL $^{\ast} \text{A}^{\ast} \text{B}^{\ast} $ color space to enhance their features. Our model was trained on a combined dataset comprising Kvasir-SEG and CVC-ClinicDB and cross-validated on CVC-ColonDB and ETIS-LaribDB. The proposed method demonstrates superior performance compared to existing methods, particularly in the detection of small and flat polyps.

Deep learning models, such as Convolutional Neural Networks (CNNs), possess the ability to learn robust features that enable effective plant disease classification. However, these models need large and diverse datasets that reflect the variations present in real-world scenarios in order to achieve high performance not only on classes that the model has already seen, but also on new classes that have not been part of the training. To address this challenge, the utilization of Conditional Generative Adversarial Network (CGAN) models offers significant advantages. CGANs can generate diverse synthetic data, thereby expanding our training dataset and enhancing the generalization capability of deep learning models. While the conventional approach typically focuses on training CGANs to generate synthetic plant disease images, our investigation goes a step further by exploring the effectiveness of training CGANs using embedded images as reference points. This research is driven by the challenge of training CGANs on high-dimensional color images, as opposed to their simpler embedding data with a low-dimension. We have discovered that CGAN trained easier in embedding space and performed better in plant disease classification for both seen and unseen compositions. Our visual analysis also showed that the small CGAN model is able to generate better synthetic embedding than the synthetic images, which leads to significant performance in the overall classification results.

Given the current expansion of the computer vision field, several applications that rely on extracting biometric information like facial gender for access control, security or marketing purposes are becoming more common.A typical gender classifier requires many training samples to learn as many distinguishable features as possible.However, collecting facial images from individuals is usually a sensitive task, and it might violate either an individual's privacy or a specific data privacy law.In order to bridge the gap between privacy and the need for many facial images for deep learning training, an artificially generated dataset of facial images is proposed.We acquire a pre-trained Style-Generative Adversarial Networks (StyleGAN) generator and use it to create a dataset of facial images.We label the images according to the observed gender using a set of criteria that differentiate the facial features of males and females apart.We use this manuallylabelled dataset to train three facial gender classifiers, a custom-designed network, and two pre-trained networks based on the Visual Geometry Group designs (VGG16) and (VGG19).We cross-validate these three classifiers on two separate datasets containing labelled images of actual subjects.For testing, we use the UTKFace and the Kaggle gender dataset.Our experimental results suggest that using a set of artificial images for training produces a comparable performance with accuracies similar to existing state-of-the-art methods, which uses actual images of individuals.The average classification accuracy of each classifier is between 94% and 95%, which is similar to existing proposed methods.

Vision Transformers (ViTs) have revolutionized medical imaging analysis, showcasing superior efficacy compared to conventional Convolutional Neural Networks (CNNs) in vital tasks such as polyp classification, detection, and segmentation. Leveraging attention mechanisms to focus on specific image regions, ViTs exhibit contextual awareness in processing visual data, culminating in robust and precise predictions, even for intricate medical images. Moreover, the inherent self-attention mechanism in Transformers accommodates varying input sizes and resolutions, granting an unprecedented flexibility absent in traditional CNNs. However, Transformers grapple with challenges like excessive memory usage and limited training parallelism due to self-attention, rendering them impractical for real-time disease detection on resource-constrained devices. In this study, we address these hurdles by investigating the integration of the recently introduced retention mechanism into polyp segmentation, introducing RetSeg, an encoder-decoder network featuring multi-head retention blocks. Drawing inspiration from Retentive Networks (RetNet), RetSeg is designed to bridge the gap between precise polyp segmentation and resource utilization, particularly tailored for colonoscopy images. We train and validate RetSeg for polyp segmentation employing two publicly available datasets: Kvasir-SEG and CVC-ClinicDB. Additionally, we showcase RetSeg's promising performance across diverse public datasets, including CVC-ColonDB, ETIS-LaribPolypDB, CVC-300, and BKAI-IGH NeoPolyp. While our work represents an early-stage exploration, further in-depth studies are imperative to advance these promising findings.

COVID-19 has beenone of the popular foci in the research community since its first outbreak in China, 2019. Radiological patternssuch as ground glass opacity (GGO) andconsolidations are often found inCT scan images ofmoderate to severe COVID-19 patients. Therefore, a deep learning model can be trained to distinguish COVID-19 patients using their CT scan images. Convolutional Neural Networks (CNNs) has been a popular choice for this type of classification task. Anotherpotential method is the use ofvisiontransformer with convolution, resulting in Convolutional Vision Transformer (ConViT), to possibly produce on par performance using less computational resources. In this study, ConViT is applied to diagnose COVID-19 cases from lung CT scan images. Particularly, we investigated the relationship of the input image pixel resolutions and the number of attention heads used in ConViT and their effects on the model’s performance.Specifically, we used 512x512, 224x224 and 128x128 pixels resolution to train the modelwith 4 (tiny), 9 (small) and 16 (base) number of attention heads used. An open access dataset consisting of 2282 COVID-19 CT images and 9776 Normal CT images from Iran is used in this study. Byusing 128x128 image pixels resolution,training using 16 attention heads, the ConViT modelhas achieved an accuracy of98.01%,sensitivity of90.83%, specificity of99.69%, positive predictive value (PPV) of95.58%, negative predictive value (NPV) of97.89%and F1-score of94.55%.The model has also achieved improvedperformance over other recent studiesthat usedthe same dataset.In conclusion, this study has shown that theConViTmodel can play a meaningful role to complement RT-PCR test on COVID-19 close contacts and patients.

Given the increased interest in utilizing artificial intelligence as an assistive tool in the medical sector, colorectal polyp detection and classification using deep learning techniques has been an active area of research in recent years. The motivation for researching this topic is that physicians miss polyps from time to time due to fatigue and lack of experience carrying out the procedure. Unidentified polyps can cause further complications and ultimately lead to colorectal cancer (CRC), one of the leading causes of cancer mortality. Although various techniques have been presented recently, several key issues, such as the lack of enough training data, white light reflection, and blur affect the performance of such methods. This paper presents a survey on recently proposed methods for detecting polyps from colonoscopy. The survey covers benchmark dataset analysis, evaluation metrics, common challenges, standard methods of building polyp detectors and a review of the latest work in the literature. We conclude this paper by providing a precise analysis of the gaps and trends discovered in the reviewed literature for future work.

Diabetes is an ever worsening problem in modern society, placing a heavy burden on healthcare systems. Due to the association between obesity and diabetes, food journaling mobile applications are an effective approach for managing and improving the outcome of diabetics. Due to the efficacy of nutritional tracking and management in managing diabetes, we implemented a deep learning-based Convolutional Neural Network food classification model to aid with food logging. The model is trained on a subset of the Food-101 and Malaysian Food 11 datasets, including web-scraped images, with a focus on food items found locally in Malaysia. In our experiments, we explore how fine-tuning of the image dataset improves the performance of the model.

The rapid increase in scientific publications makes it difficult for researchers to keep up with the latest literature and to explore new research directions. The literature-based discovery (LBD) systems aim to resolve this issue by bridging literatures from disparate fields to assist researchers in knowledge discovery and the formulation and testing of research hypotheses. Previous studies have focused mainly on evaluating the efficacy of LBD systems by replicating historical LBD events. The usability of LBD systems has been under-researched, which partly explains the low adoption of the systems. This paper presents a survey study that evaluates the usability of a LBD system for knowledge discovery and hypothesis refinement, and also investigates factors affecting its adoption among biomedical researchers in Sarawak, Malaysia. The findings suggest that the adoption of the LBD system is related to their perceived usefulness and perceived difficulty in interacting with the user interface features of the system.

This research attempts to construct a picture of the current professional situation concerning educators in Autism Spectrum Disorder (ASD) amelioration in Sarawak, Malaysia. Data was collected through interviewing ASD educators from licenced special needs education establishments, then analysed for recurring trends. Findings revealed that although there is a visible need for educators, there appears to be very little in the way of guidelines, recognition, support, and career opportunities for these practitioners. Backing from a literature review suggests that an industry standard in terms of remuneration, certification, and a regulatory body may assist with educator retention and training. However, difficulty in securing opportunities for professional self-improvement in this field has resulted in unique and divergent educator development, demonstrated by a history of self-reliance and adaptability as service providers due to necessity. This study examines the gaps in service for ASD individuals through the experiences of ASD educators in Sarawak. It highlights the distinct need for more research in the field for this locality in order to pinpoint and amend its weaknesses. Further study would be valuable for not only the developing urban areas here, but due to shared cultural aspects, also have implications for rural areas and surrounding developing nations.

COVID-19 has been raging for almost three years ever since its first outbreak. It is without a doubt that it is a common human goal to end the pandemic and how it was before it started. Many efforts have been made to work toward this goal. In computer vision, works have been done to aid medical professionals into faster and more effective procedures when dealing with the disease. For example, disease diagnosis and severity prediction using chest imaging. At the same time, vision transformer is introduced and quickly stormed its way into one of the best deep learning models ever developed due to its ability to achieve good performance while being resources friendly. In this study, we investigated the performance of ViT on COVID19 severity classification using an open-source CXR images dataset. We applied different augmentation and transformation techniques to the dataset to see ViT’s ability to learn the features of the different severity levels of the disease. It is concluded that training ViT using the horizontally flipped images added to the original dataset gives the best overall accuracy of 0.862. To achieve explainability, we have also applied Grad-CAM to the best performing model to make sure it is looking at relevant region of the CXR image upon predicting the class label.

Automatic facial gender prediction is an essential sub-task of any facial analysis system. Given the current rapid advancements of artificial intelligence and the ease of connectivity, gender prediction models are being utilised in many applications, such as soft biometrics, surveillance, targeted marketing, social media, human-computer interaction applications and many more. Nonetheless, when gender prediction models attempt to classify images taken in unrestrained conditions with different poses and illuminations, the classification accuracy drops significantly. One solution to solve this issue is employing a very deep architecture to extract as many features as possible from an image. However, such models are usually complex to train and might not be deployable to devices with limited computation power. Therefore, the main objective of this study is to close the gap between the classification of unrestrained facial images and the need for an efficient architecture. We achieve this objective by presenting a novel lightweight deep convolutional neural network to classify male and female facial images taken in uncontrolled environments. We train our model on the UTKFace dataset and produce competitive accuracies compared to several state-of-the-art methods. We then carefully analyse and discuss our results to better understand the variables that control the performance of our model. In addition, we present potential suggestions and ideas to enhance our proposed method for future research.