Theses - Undergraduate

Browse

Search Results

Now showing 1 - 10 of 384
  • Thumbnail Image
    Item
    MediGenius AI:Fracture Recovery Revolutionized Using AI
    (North South University, 2023) Md. Khurshid Jahan; Ashrin Mobashira Shifa; Rokaeya Sharmin; DR. MOHAMMAD ASHRAFUZZAMAN KHAN
    This study introduces an innovative approach to fracture recovery utilizing artificial intelligence (AI) technology. A dataset comprising 1012 X-ray images, including 331 instances of fractures, sourced from St. Mariyaam Diagnostic Center, was meticulously annotated for nine distinct fracture types. Collaborative efforts with Dr. Rakibul Islam and Md. Asiful Rahman Maruf ensured accurate labeling and enriched the dataset with corresponding NLP descriptions and patient adviceAnticipating future enhancements, the integration of diffusion models is proposed, with the aim of synthesizing high-fidelity X-ray images. This development holds substantial promise in redefining fracture recovery procedures.
  • Thumbnail Image
    Item
    Prostate Cancer Cell Prediction from Histopathological Images using Convolutional Neural Network
    (North South University, 2022) Watan Al Arafat; Md. Mushfiqur Rahman; Dr. Mohammad Monirujjaman Khan
    In cancer research, pinpointing a patient's future response to treatment is crucial for making informed decisions. This study delves into a potential method for predicting the return of prostate cancer after surgery, utilizing imagery from tissue samples. Scientists analyzed a group of patients, categorizing them based on whether their cancer returned after treatment. To account for existing variations besides the cancer itself, they meticulously matched patients within the group based on factors like age, ethnicity, and disease severity. The proposed technique hinges on a sophisticated form of artificial intelligence known as deep learning. Uniquely, it employs two distinct AI models: one to pinpoint individual cells within the tissue, even in dense areas, and another to classify these cells. By analyzing these classified cells, the researchers were able to estimate, with promising accuracy, the likelihood of cancer recurrence in a patient. This method, if further validated, holds the potential to revolutionize treatment decisions for prostate cancer, offering a more personalized approach. The broader implications of this research extend beyond prostate cancer. This approach, with further development, might be adaptable to predicting treatment outcomes in various cancers. We have found the accuracy 93.33 percent. Keywords- Prostate; Cancer; Convolutional Neural Network; CNN; Deep Learning, prediction
  • Thumbnail Image
    Item
    Cardiovascular Disease Detection using Transfer Learning Approach
    (North South University, 2024) Jahangir Alam; Tashdid Alam; Hafsa Tasnim Badhan; Dr. Shahnewaz Siddique
    Cardiovascular Disease (CVD) is a multidimensional worldwide health concern, covering illnesses such as myocardial infarctions, cerebrovascular accidents and excessive blood pressure. Understanding its risk factors is crucial for successful preventative actions and therapies. This research is inspired by the astounding effect of CVD, responsible for 17.9 million deaths yearly, comprising 31% of global mortality [32]. Encouragingly, the American Heart Association says up to 80% of CVD cases are avoidable by lifestyle adjustments and early therapies. Proactive management of risk factors is crucial, with the healthcare sector, equipped with significant patient data, playing a critical role. The paper presents a comprehensive investigation into the development and evaluation of deep learning models for binary classification in cardiovascular disease detection. Four distinct models like ResNet50, EfficientNetB0, CNN and VGG16 were explored, each utilizing different architectures and transfer learning techniques. The study begins with a detailed description of data loading and preprocessing procedures, highlighting the importance of standardized image preparation techniques for effective model training. Each model's architecture is thoroughly discussed, elucidating the unique characteristics and design choices that contribute to their performance. The ResNet50 model employs residual blocks to address the vanishing gradient problem, while the EfficientNetB0 leverages transfer learning with pre-trained layers from ImageNet. The CNN model features a basic yet successful architecture with convolutional and dense layers, whereas the VGG16 model adopts a deep and uniform structure with 3x3 convolutional kernels [33]. Training procedures and results are meticulously documented for each model, encompassing aspects such as epoch-based training, batch sizes and optimizer selection. Performance metrics including accuracy and loss are analyzed across epochs to assess model convergence and generalization capabilities. Notably, the EfficientNetB7 model outperforms others, achieving remarkable accuracy and minimal loss on both training and validation datasets, demonstrating its efficacy in cardiovascular disease detection. The ResNet50 model achieved a training accuracy of 86.39% and a validation accuracy of 84.72%. The EfficientNetB0 model attained a training accuracy of 97.32% and a validation accuracy of 98.83%. The CNN model demonstrated a training accuracy of 80.55% and a validation accuracy of 87.83%. Finally, the VGG16 model achieved a training accuracy of 95.13% and a validation accuracy of 99.55%. The paper also presents the development of a user-friendly web application using Streamlit, facilitating easy access to the trained models for real-world application [34]. The application comprises instructional, classifier and about us pages, each designed with clarity and simplicity to enhance user experience. Sample images and immediate classification results contribute to the usability and trustworthiness of the application, catering to medical professionals' needs for accurate and timely diagnostics [34]. The study provides valuable insights into the application of deep learning models for cardiovascular disease detection, elucidating the significance of architecture selection, data preprocessing and training procedures. The findings underscore the efficacy of transfer learning techniques and model architectures in achieving high accuracy and reliability in medical image classification tasks. The developed web application offers a practical solution for real-world deployment, facilitating seamless integration of deep learning technologies into clinical workflows.
  • Thumbnail Image
    Item
    Recommendation System by Analyzing User Review Using Machine Learning
    (North South Universty, 2021) Md. Iqbal Hossain; Md. Rahat Hossain; Fariha Ahmed Raha; Mohammad Monirujjaman Khan
    This report presents the design and the implementation of Recommendation System by analyzing user reviews using machine learning. The system will filter using hybrid filtering (content based and collaborative) for users’ choices and analyzing reviews of users for that product, more items are recommended. It merged some categories of items from the dataset and used Truncated SVD (Singular value decomposition) to reduce the number of unnecessary features and dimensions of the dataset. It made a correlation between all the items with items purchased by the user based on items rated by other users who bought the same products. The algorithms are SVD, KNN and KNNWithMeans which has been applied to the system to make a model for recommendation. Overall, the system produced satisfactory results, successfully recommending top 10 products to users having the same taste of choice while also maintaining its design objectives of ring low cost and user friendly.
  • Thumbnail Image
    Item
    RECK: IOT- Based Fire Alarming System and Help Service
    (North South University, 2019-12-31) Mahamudhul Hasan; Md Shahriar Karim
  • Thumbnail Image
    Item
    IoT Based Automated Hydroponic Farming System
    (North South University, 2021-12-31) Sakib Asrar; Fahim Tanzil Takin; Ihfaz Tahmid Morshed; Tanvirul Azim; Dr. Sifat Momen
    The objective of this project is to control the environmental elements for farming using automated technology. An IoT based automated hydroponic system is proposed by which crops can grow in water with the necessary elements of the soil. The hydroponic method reduces water consumption up to 95% of regular farming, automates the regulation of the environment and nutrients in the farm to maximize production. Therefore, farmers can control all environmental elements with a minimum of time and effort. The proposed system can be remotely controlled and monitored via IoT. Using this farming method, crops can grow in any environment regardless of places and seasons.
  • Thumbnail Image
    Item
    Distinguish walking and running
    (North South University, 2021-12-31) Sangeeta Paul Priya; Dr. Md Shahriar Karim
    Physical activity is a vital need for our survival. But in the busy world that we live in, it’s not often that easy for us to include exercise within our schedule. But the least we can do is keep track of how much physical activity we are doing throughout the day that can actually leave an impact on our physic. Running impact us differently than just walking. That’s why it is important to keep track of both individually. It becomes even more important especially if we go out to run with the intention of exercising. That’s why our goal is to create a system that can learn the difference between running and walking from data through machine learning and provides the user with a result that contains how much they ran and how much they walked separately. Many systems were developed over the years to distinguish walking and running. This project is about establishing a system that can detect whether someone is running or walking. For this project we went with two different approaches, both of which involved machine learning. Our first approach was to use a numeric dataset and apply them on different machine learning algorithms. Our other approach was to use an image-based dataset to create a CNN model. The end goal for both processes was to interphase the models with an Arduino.
  • Thumbnail Image
    Item
    Fake Job Posting Detection Using Machine Learning
    (North South University, 2022-12-30) Tawfiqul Islam Talukder; S.M Shouvik Islam; Dr. Mohammad Monirujjaman Khan
    This report presents the design and the implementation of a system that can detect fake jobs using a machine learning method that employs a variety of categorization algorithms. The COVID-19 epidemic situation has transformed the regular livelihoods of mankind in the world. This epidemic has put excessive pressure on the job market. As a consequence of the epidemic, most organizations have halted their recruiting processes, which has raised the rate of unemployment. Online recruiting has suddenly increased the quantity of applicants while also bridging the distance between recruiters and candidates. It indicates that scammers have emerged in the online recruiting market. They provide extremely high pay ranges or any other type of benefit on several online platforms. It's called "Fake Job Postings." Job seekers are applying for those fake jobs. As a result, scammers steal their personal information. Scammers use their personal information for a variety of cybercrimes or sell it on the dark web. This paper's objective is to identify and verify these job advertisements, whether they’re fake or not. To identify these fake job advertisements, Machine Learning Algorithms (MLA) was implemented throughout this study, such as the Random Forest algorithm, and Logistic Regression algorithm. This study trained and tested the dataset and got an accuracy of 98.86 percent in Logistic Regression and 98.54 percent in Random Forest. In the Logistic Regression algorithm, our recommended technique has an accuracy of 98.86 percent, which is a huge improvement over the current methods. The accuracy percentile of both the algorithms used throughout this analysis is substantially in excess of prior studies, showing that the algorithms utilized throughout this analysis are well balanced.
  • Thumbnail Image
    Item
    Speaking System for Mute and Deaf people
    (North South University, 2021-04-30) Muhtasim Rafid Ahmed; Fareeza Sharara karim Bohota; Atik Mahmud; Dr. ATIQUR RAHMAN
    Deaf and mute communities are facing big problem for their disability. They are not comfortable with normal people. So, the aim of our project is to eradicate the barrier between disability and normal people in terms of communication. We have made a simple, wearable sensor-based project which is low cost and anyone can easily wear this. Our project is smart hand gloves and portable, and using flex sensor, amplifier, Arduinouno, Arduinolilypad, speaker, LCD screen, SD card. Through our project speaker can talk for mute and blind people and also deaf people can see the LCD screen. Standard ASL hand gesture taking as input database. ASL is standard sign language invented for mute and deaf people so that they are live their life as normal people. They also can work in different sector using our project. We did not use image processing or PIC microcontroller only because of high cost. We tried to make this project in low cost so that everyone can use this
  • Thumbnail Image
    Item
    Health care chat-bot hospital management system
    (North South University, 2020-08-30) MD.Harun ur Rashid.; MD. Maruf Hasan; Yeasin Arafat Prantik; Dr. ATIQUR RAHMAN
    Our main focus is design a unique Healthcare chat-bot Hospital Management System that will improve hospital experience for both patients and the hospital authorities. The whole system will run on internet. The system is written in PHP, java script, j query, HTML and CSS. Users will have the felicity to log in from any place with internet connection. After that they will be able to various tasks that are designed for them. Users are categorized in three groups :( Management, Patient and Doctor).The primary target is to focus on every user who can get our service and get benefited. It can be turned into a paid system only for doctors.Where the doctors can get additional cloud storage on payment. A doctor can have different types of patient and the number of patients also vary from doctor to doctor. A doctor can have various number of patients. We can assume that doctors will need different amount of cloud storage. We can allocate a fixed cloud storage for each doctors. They can ask for extra storage according their demand and they will be charged for their demand. We can make various package and assign various cost. The patients will have some allocated space which they can use to keep their information. As the patient only needs storage for only themselves they can use this as a free user. This will make the system useful and more convenient for everyone. Keeping the goal in mind the system we developed works as a social network where information’s are more close and relevant for every user.This report contains the full details of the system and its functionality in details