http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/5543 2026-06-28T02:00:35Z 2026-06-28T02:00:35Z Nahid, Beniamine Al http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17274 2026-06-10T21:02:19Z 2025-01-15T00:00:00Z

Building Trust In Ai-Driven Skin Disease Diagnosis Through Explainable Ai Nahid, Beniamine Al This research focuses on developing an AI-based system for skin disease diagnosis that achieves high accuracy while ensuring interpretability and transparency. The proposed system leverages ResNet101 as the backbone model and achieved an impressive 98% accuracy across six skin disease categories: Acne, Carcinoma, Eczema, Keratosis, Milia, and Rosacea. To address the critical challenges of trust and usability in clinical settings, Explainable AI (XAI) techniques, such as LIME, were integrated. These techniques provide detailed visualizations of class-specific probabilities and regional contributions, enabling both patients and dermatologists to better understand and trust the model’s predictions. Extensive experiments were conducted, comparing the performance of ResNet101 against other pre-trained models, including VGG16, ResNet50, and EfficientNetB7. The results highlight the superior feature extraction capabilities and generalization performance of ResNet101, which outperformed other models in accuracy, precision, recall, and F1-score. This research underscores the importance of combining technical accuracy with explainability to enhance trust in AI systems, thereby supporting patient-centered care. By addressing the gap between advanced AI technology and practical healthcare applications, this study contributes to the broad-scale adoption of reliable and transparent AI systems in dermatology and other medical fields. Thesis Report

2025-01-15T00:00:00Z Eyashin, Md. http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17273 2026-06-10T21:02:28Z 2025-01-11T00:00:00Z

A Colab-Integrated Hybrid Model For Effective Ransomware Detection Via Voting Classifier Techniques Eyashin, Md. Ransomware is a harmful malware that is designed to encrypt a victim's data or lock the victim's system, then demand a ransom for restoration or decrypting the data or unlocking the system, and it often causes significant financial and operational damage. Current ransomware detection methods are struggling to detect ransomware properly because most of the ransomware detection approaches follow dynamic analysis techniques which involve a complicated process, and also use only signature-based features, not use network or behavioral based features. This study proposed a ransomware detection hybrid model that is based on static analysis and uses signature- based features, network, or behavioral features. This study used three ML models for implementing hybrid models, models are Decision Tree, Random Forest, and K-Nearest Neighbors. This study proposed two hybrid models, where the hybrid model achieved highest detection accuracy 97.48% with a low false positive and false negative rate. Thesis Report

2025-01-11T00:00:00Z Arafat, Md. Yeasin http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17272 2026-06-10T21:02:13Z 2025-01-15T00:00:00Z

Automating Clinical Note Summarization Using LLM Arafat, Md. Yeasin The increasing volume of electronic health records (EHR) and the growing complexity of clinical documentation have highlighted the need for efficient and reliable summarization tools. This thesis explores the application of large language models (LLMs) in automating clinical note summarization, with the goal of reducing the cognitive and administrative burden on healthcare professionals while maintaining clinical accuracy and relevance. Leveraging state-of-the-art LLM architectures such as T5 and FLAN-T5, this research focuses on extracting key information from clinical notes, including patient diagnoses, treatment plans, and medical histories, and generating concise, structured summaries suitable for clinical workflows. The study evaluates the performance of fine-tuned LLMs on datasets such as MIMIC-III, MeQSum, and ProbSum using quantitative metrics like ROUGE and BERTScore, achieving a ROUGE F1 score of 0.95 and demonstrating high efficiency with a runtime of 2.35 seconds per note. Qualitative analysis confirms the generated summaries' clinical relevance, with outputs aligned to standard sections like diagnoses, imaging results, and treatments. Despite these strengths, challenges such as occasional hallucinated information, omitted secondary details, and inconsistent formatting are identified. Results from physician feedback underscore the practicality of LLMs in improving healthcare documentation, with models like LLaMA-Clinic achieving over 90% acceptance in a blinded review. Additionally, cost analysis reveals a 3.75-fold reduction in inference costs compared to proprietary alternatives, emphasizing the feasibility of open-source solutions. This research highlights the potential of LLMs to enhance clinical workflows, reduce diagnostic errors, and improve patient care. Future directions include expanding datasets for better generalizability, addressing hallucination issues, and ensuring seamless integration into healthcare systems through ethical and clinician-centered approaches. The findings reinforce the role of AI in advancing healthcare, promoting accessibility, and addressing global challenges in medical documentation and decision-making. Thesis Report

2025-01-15T00:00:00Z Turja, K.M. Al Jaziz http://dspace.daffodilvarsity.edu.bd:8080/handle/123456789/17271 2026-06-10T21:02:10Z 2025-01-15T00:00:00Z

High-Precision 3D Spleen Segmentation for Surgical Planning Using Deep Learning Turja, K.M. Al Jaziz Medical imaging has become an indispensable tool in modern healthcare, facilitating accurate visualization of internal anatomical structures for diagnosis, treatment planning, and surgical procedures. Among these, segmentation of organs such as the spleen is particularly critical for pre- surgical workflows, including radiation therapy planning, volume estimation, and surgical navigation. The spleen, an organ of considerable clinical significance, often requires precise delineation for managing trauma or pathologies like lymphoma. However, spleen segmentation from computed tomography (CT) images is challenging due to anatomical variability, low tissue contrast, and the presence of imaging artifacts. Traditional segmentation methods, such as thresholding and active contour models, often fail to generalize across diverse datasets, necessitating advanced approaches. This study aims to address these challenges by developing a robust segmentation framework based on a 3D U-Net architecture. The primary objectives include achieving high segmentation performance on the Medical Segmentation Decathlon (MSD) Task 09 spleen dataset, evaluating the model's generalization capabilities, and demonstrating its clinical relevance for pre-surgical workflows. The methodology involves a systematic preprocessing pipeline to normalize and augment the CT volumes, followed by training a 3D U-Net model using Dice loss and the Adam optimizer over 600 epochs. The segmentation performance was evaluated using metrics such as Dice similarity coefficient and validation loss to ensure accuracy and reliability. The proposed framework achieved remarkable results, with a training Dice similarity coefficient of 0.9582, a validation Dice similarity coefficient of 0.9494 and a test Dice Similarity coefficient of 0.9483. These results highlight the model's strong generalization ability and effectiveness in addressing challenges such as noisy imaging and anatomical variability. Qualitative evaluations further confirm the precision of the segmentation, with predicted masks showing high alignment with ground truth annotations. In conclusion, this study presents a significant contribution to automated medical image segmentation by leveraging the power of deep learning. The results not only validate the efficacy of the proposed framework but also lay the groundwork for extending this approach to multi-organ segmentation tasks and real-time clinical workflows. Future research directions include integrating multi-modal imaging data, improving computational efficiency, and validating the framework in clinical settings to ensure broader applicability and impact. Thesis Report

2025-01-15T00:00:00Z