Research Projects

 Deep Learning for Computed Tomography Scans Analysis the swinUNETR and nnUNet models show promising results for abdominal organ segmentation in CT scans, achieving near state-of-the-art performance on WORD and AMOS data. Introduces deep learning, explores architectures like U-Net and Swin UNETR, and compares nnU-Net with Swin-UNet [View] 

MediTandem: Enhancing Small Medical Imaging Object Segmentation through Classification-based Refinement develop a novel approach that enhances small medical object segmentation by integrating a classification-based refinement. Evaluated on 1,076 annotated kidney tubule images, it outperforms standard models, achieving 82.4% IoU and improving cyst-wise Precision from 72.3% to 77.6%.[View] 

NMRGen: Generative Modeling Framework for Molecular Structure Prediction from NMR Spectra Introduce a generative framework using a SMILES autoencoder (GRU-based) and an NMR encoder (CNN+DNN) to predict molecular structures from NMR spectra. While the SMILES autoencoder performed well, the NMR encoder struggled, resulting in 72% accuracy but low SMILES validity and few valid molecular predictions. [View] 

 A Neural Networks Based Heartbeat Classifications using ECG Signals uses deep convolutional neural networks and transfer learning, this approach classifies ECG beats into normal and abnormal. The models show strong performance on specific patients' data, with individual classifiers achieving a balanced accuracy of 94.6%. [View]

 IoT Based Industrial Water Pollution Evaluation System final year project used  pH, TDS, Water Temperature and Turbidity sensors connected to an ESP32 microcontroller. The collected data was stored in an Android application, providing a comprehensive solution for monitoring water quality in real-time.[View]

 Self-Supervised Learning for Zero-Shot Pediatric Tuberculosis Detection in Chest X-rays A novel approach using self-supervised learning within Vision Transformers (ViT) is proposed to improve TB detection in chest X-rays, particularly for zero-shot pediatric cases. This approach reduces reliance on large annotated datasets and improves detection performance, but faces limitations in generalizability and ethical implications. [View]

 Magnetic Resonance Imaging (MRI) to Enhance Brain Tumor Recognition  In this project, I used CNNs and Python libraries such as NumPy, Pandas, and SciPy. I also employed Jupyter Notebook for data exploration and model development. The model results demonstrated a validation accuracy of 91% and a test set accuracy of 89%. [View] 

 SUBMIP  is a cloud-based platform that uses wearable sensors, AI, and predictive analytics to provide real-time health monitoring and insights, revolutionizing healthcare by enabling early disease detection, personalized health recommendations, and improved quality of life. [View]

 Crop Disease Detection Using Deep Learning Computer Vision Deep learning computer vision is used for crop disease detection, enhancing early detection and management, improving agricultural productivity and sustainability. It uses CNNs for image classification, preprocessing techniques, high-quality annotated images, TensorFlow and PyTorch training frameworks, and real-time deployment. [View]

 Emotion Detection using Machine Learning my 3rd-year project used Python libraries like NumPy, Pandas, TensorFlow, and Scikit Learn to implement a deep learning model for image processing, enhancing feature extraction and classification. [View]

 Plant Disease Detection Based on Image Processing This project aims to develop a system that can detect plant diseases by analyzing images of leaves. The system uses image processing techniques and machine learning algorithms to classify images of diseased and healthy plants with high accuracy. The project uses Python, OpenCV, TensorFlow, Scikit-learn, and various tools for data collection, preprocessing, feature extraction, and deployment. [View]

 Machine Learning based Disease Prediction Python libraries, machine learning algorithms, healthcare datasets, advanced techniques, ethical collaboration and secure web frameworks like Flask. Ensure ethical considerations, collaborate with healthcare professionals and prioritize privacy and security. [View]

 Water4.0: An Industrial Water Pollution Forecasting Using Machine Learning the KNN model outperforms traditional methods in industrial water pollution forecasting, achieving 91.5% accuracy. Its precision of 93.2% and recall of 90.1% minimize false positives, making it a valuable tool for various industries, ensuring environmental compliance and effective proactive measures. [View]