Milestone Transactions on Medical Technometrics

Machine Learning Technique for Brain Tumor Detection: A Comprehensive Review

Kamleshwar Dubey — 2025-02-03

Detecting

diseases in the brain at early stages or proper diagnosis and treatment of tumor in the brain are perhaps the biggest challenges facing medical sciences. New advancement in Machine Learning has developed novel methods for augmenting accuracy and effectiveness of detecting brain tumors based on image obtained through MRI scans. This review goes in-depth analysis of how detection and classification is currently improving, highlighting such techniques involving SVM, CNNs, ResNet, and Xception. Key spatial and textural feature extraction techniques, like PCA, HOG, and GLCM, have also been considered in order to improve the classification performance. Ensemble approaches, such as Random Forest, have been taken into consideration for handling the complexity of the dataset. The study evaluates the benefits, limitations, and clinical applicability of these methods in view of the challenges posed by diverse datasets, high computational demands, and generalization issues. Further research and development are required to make the machine-learning diagnostic capabilities of brain-related conditions more reliable and practical.

Automated Tuberculosis Detection from Chest X-Rays Using a ResNet50 Architecture

Sreelakshmi Induri — 2025-02-21

TB (tuberculosis) is still a major worldwide health concern, and in order to control its spread and enhance patient outcomes, precise and prompt diagnostic methods are required. This study introduces a deep-learning methodology employing a ResNet50 architecture for the reliable and precise classification of TB in CXR images. The model utilizes a robust CNN (Convolutional Neural Network) to extract essential features, enhanced by image pre-processing techniques to optimize image quality and overall performance. The ResNet50 classifier shows great performance metrics, achieving an accuracy of 99.82%, a 99.82% accuracy rate and an AUC of 99.87%. These results highlight the potential of ResNet50 as a valuable TB detection tool for healthcare professionals, particularly in resource-constrained environments, enabling faster diagnosis and treatment.

Deep Learning-Based Detection of Hair and Scalp Diseases Using CNN and Image Processing

Busireddy Seshakagari Haranadha Reddy — 2025-03-04

Hair and scalp disorders affect millions worldwide, often undiagnosed in their early stages due to similarities with normal hair shedding. Conditions such as alopecia, psoriasis, and folliculitis require professional dermatological evaluation, which can be time-consuming and lead to delayed treatment. Early diagnosis and clinical decision-making have been strengthened by automating disease detection in healthcare through the development of deep learning and improved image-processing techniques. This study evaluated the three primary conditions associated with the scalp using a convolutional neural network (CNN) model. The research faced challenges due to the restricted availability of structured datasets and variations in image quality from several sources. To address this, 150 images were gathered from various repositories and subjected to pre-processing techniques, including denoising, contrast enhancement, image equalization, and data balancing. The processed dataset was employed to train a 2D CNN model with a training and validation accuracy of 97% and 92%. Additionally, we curated a dataset of scalp images to support further research in this domain, promoting the development of more robust automated diagnostic systems.

Deep Learning-Based Identification of High-Risk Zones for Lumpy Skin Disease in Cattle

G Ramasubba Reddy — 2025-03-11

Lumpy skin disease (LSD) is a virus infection in cattle, mainly transmitted through mosquitoes, such as blood-eating insects. The disease creates essential challenges for the livestock industry by affecting milk and beef production and domestic and international trade. Resource limitations, lack of specialized expertise, and lack of time often obstruct traditional clinical approaches. This study introduces a deep learning-based framework for segmenting and classifying LSD-affected skin regions to address these challenges. The proposed approach leverages a 10-layer Convolutional Neural Network (CNN) trained on a curated Cattle Lumpy Skin Disease (CLSD) dataset. The significance of skin discoloration in disease identification is that a color histogram enhances feature extraction. The segmented affected regions undergo further processing through a deep pre-trained CNN for feature extraction, followed by threshold-based binarization. Finally, classification is performed using an Extreme Learning Machine (ELM) classifier, achieving an accuracy of 96% on the CLSD dataset. Comparative analysis with existing state-of-the-art techniques demonstrates the effectiveness of the proposed methodology, highlighting its potential for reliable LSD diagnosis in cattle.