Glaucoma Detection Based on Fundus Images using Machine Learning Techniques
An Artificial Intelligence Approach to Lung Cancer Diagnosis using LungNet-TL Model
Optimized Hybrid Model Combining Hidden Markov and Stochastic Neural Networks for Groundwater Quality Forecasting in Kanyakumari, Tamil Nadu, India
Second-Order Polynomial Interpolation Filters for Image Demosaicking with Perceptual-Based Tone Mapping and Quantum-Inspired Optimization
E-Commerce Product Recommendation using Machine Learning Techniques
Glaucoma is a leading cause of irreversible blindness worldwide, frequently progressing without noticeable symptoms until significant vision loss occurs. Early detection is essential for managing and mitigating its effects. However, traditional diagnostic methods, such as intraocular pressure tests and optical coherence tomography (OCT), may not be easily accessible in resource-limited areas. Fundus imaging, which captures the back of the eye, provides a non-invasive method to assess the optic nerve and retinal health, making it an effective option for glaucoma screening. By leveraging large datasets of labeled fundus images, machine learning algorithms can be trained to recognize early structural changes in the optic nerve and surrounding retinal nerve fiber layers indicative of glaucoma. Convolutional Neural Networks (CNNs) are especially useful in this application, as they can automatically extract relevant features from complex images without requiring manual intervention.This paper explores a machine learning-based approach for glaucoma detection using fundus images, aiming to develop an accessible and efficient diagnostic tool. This paper proposed Sobeledge detection for data preprocessing, Convolutional Neural Networks (CNNs) for model selection, training, and evaluation. The proposed approach has the potential to provide accurate and scalable glaucoma screening solutions, potentially aiding early diagnosis and reducing the burden on healthcare systems. This method attains better accuracy rate.
Lung cancer is one of the deadliest types of cancer worldwide, and early diagnosis is critical for improving patient outcomes. This study proposes LungNet-LT, a novel deep learning model specifically designed for detecting and classifying lung cancer using medical imaging data. The LungNet-LT model enhances feature extraction from computed tomography (CT) scans and X-ray images by integrating convolutional neural networks (CNNs) with transfer learning techniques. Compared to conventional diagnostic methods, LungNet-LT demonstrates substantial improvements in classification sensitivity, specificity, and accuracy by leveraging a pre-trained model fine-tuned on lung-specific datasets. Advanced image processing techniques are incorporated to reduce noise and improve cancer cell localization, enabling more reliable predictions for both early-stage and advanced lung cancer cases. The performance of LungNet-LT is validated on publicly available lung cancer datasets and a curated clinical dataset from a partner hospital. Results indicate that LungNet-LT achieves a diagnostic accuracy of over 98.9%, with a significant reduction in false positives and false negatives. These findings highlight the potential of LungNet-LT as a powerful tool to assist physicians in diagnosing lung cancer and improving patient outcomes through prompt and precise actions.
Groundwater quality is an important environmental concern, particularly in areas like Kanyakumari District, Tamil Nadu, where there is heavy reliance on groundwater for drinking and agriculture. To manage underground resources sustainably, understanding groundwater quality parameters is essential. In this study, a novel hybrid model that combines the strengths of the Hidden Markov Model (HMM) and Stochastic Neural Network (SNN) is employed to predict the quality of groundwater in the study area with high accuracy. Utilizing groundwater quality indicators such as pH, EC, TDS, major ions, and minor ions, along with rainfall data (since it is one of the factors influencing groundwater quality in the study area), the HMM captures latent sequential patterns within the dataset, transforming the features to enhance SNN classification. The model's performance is further optimized, achieving an overall accuracy of 97%. Additionally, the confusion matrix, classification report, Cohen Kappa score, and Matthews correlation coefficient are used to assess the model's performance. The training and testing accuracy are used to evaluate the generalization of unseen data. This study contributes to the development of advanced tools for groundwater management.
An innovative method for image demosaicking is introduced, leveraging perceptual-based tone mapping, secondorder polynomial interpolation filters, and Quantum-Inspired Optimization to achieve superior image reconstruction quality. The proposed approach integrates an Autoregressive Wavelet Water Optimization (WWO) algorithm to determine coefficients for second-order polynomial filters within the LPA-ICI framework. Simultaneously, a Deep Convolutional Neural Network (Deep CNN) is employed to generate residual images, capturing intricate features. The outputs of the interpolation-based method and the Deep CNN are fused using an entropy-based metric, resulting in enhanced visual quality and reduced artifacts in the demosaicked images. Perceptual-based tone mapping is applied to address brightness discrepancies, ensuring luminance accuracy and improved image realism. Additionally, Quantum-Inspired Optimization enhances the efficiency and robustness of the filtering process. Experimental results demonstrate significant improvements in reconstruction accuracy, making the proposed method a promising alternative for applications requiring precise and visually appealing demosaicking. Future work will explore the extension of this method to multispectral images and address the challenges of real-time processing.
Machine learning is progressively being adopted by e-commerce platforms to enhance the shopping experience for consumers. By utilizing machine learning, large user datasets are analyzed to effectively forecast customer preferences, allowing for more relevant and personalized recommendations. Techniques such as collaborative filtering predict interests based on groups of similar users, while clustering, or segmentation, is employed for both users and items. This approach helps mitigate issues related to data sparsity and the cold-start challenge when it comes to generating valuable recommendations. Representation learning, particularly through deep neural networks, can capture complex patterns, which lead to high-quality recommendations. Additionally, LightGBM has shown enhancements in performance efficiency and its ability to manage very large data sets effectively. Hybrid models combine collaborative filtering with content-based filtering to achieve greater precision in recommendations. This review discusses cutting-edge ecommerce recommendation systems and how these advanced machine learning strategies work together to improve customer satisfaction, drive sales growth, and enhance competitiveness in the evolving e-commerce landscape.
