Department of Computer Science

Permanent URI for this communityhttps://repository.cuilahore.edu.pk/handle/123456789/16

Browse

Author: search.filters.author.Dr. Muhammad Aksam IftikharSubject: search.filters.subject.Dr. Muhammad Aksam Iftikhar

Search Results

Now showing 1 - 7 of 7
  • No Thumbnail Available
    Item
    Enhancing Brain Tumor Diagnosis through Deep Feature Learning and Spatial Context Awareness in MRI Scans
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2025) Qasim Ali; FA22-RCS-022; Dr. Muhammad Aksam Iftikhar; LHR TP 9488
    Brain tumor classification is a critical task in medical diagnostics, where accurate and timely detection significantly influences treatment planning and patient outcomes. In this study, a hybrid deep learning framework which joins Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) is developed to classify brain tumors based on the BRATS 2021 dataset using the FLAIR modality. The CNN part learns local patterns (e.g. edges and textures) while the ViT part captures global dependencies over the whole images, which provides insights of the tumor regions in a holistic way. After performing pre-processing on the dataset, which comprised of 53,610 images, the images were partitioned into tumor and non-tumor groups. The training, validation, and test sets were created with 80%, 10%, and 10%, respectively. The listed model performed two independently trained models for 100 epochs at first, then is applied a feature fusion technique which exploits the complementarity of both architectures. The hybrid model was assessed based on conventional metrics like precision, recall, F1-score, and accuracy. The result confirmed that the hybrid method achieved better performance than the individual CNN and ViT models, with improvements in precision and strengthened classification capability. The study emphasizes how hybrid deep learning techniques can improve clinical applications by alleviating challenges related to data variability, noise, and image complexity. The proposed method combining local and global feature extraction methods ensures valuable typing providing a reliable and scalable implementation of automatic brain tumor classification helping improve the accuracy and efficiency of healthcare related problem solutions and medical diagnostics.
  • No Thumbnail Available
    Item
    Self-Attention Based Multi-Model Fusion Mechanism for Diabetic Retinopathy Grading
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2024) Sana Naseem; FA22-RCS-023; Dr. Muhammad Aksam Iftikhar; LHR TP 9332
    Among the five basic human senses, vision is the most versatile, which depends primarily on the flawless working of human eyes. An eye condition called diabetic retinopathy (DR) is caused by diabetes. Retinal lesions are important signs of DR, and they can be used for early diagnosis. This research will identify and evaluate any pathological alterations brought on by the development of DR. In this research, an attention based deep learning method is proposed to classify Diabetic Retinopathy (DR), wherein CLAHE preprocessing is first used to enhance the image contrast. The publicly accessible dataset APTOS 2019 is utilized in the experiments to evaluate the effectiveness of proposed methodology. However, this dataset has imbalanced class distribution, which was balanced using different data augmentation strategies. For effective feature learning, four pre-trained models namely Inception v3, VGG-19, ResNet-50, and DensNet-121 are modified by adding self- attention blocks following each convolutional base of the pre-trained models. These self-attention blocks enable the model to selectively focus on important parts of DR lesions. Subsequently, the extracted features are fused using late fusion, and a Multi- Layer Perceptron (MLP) classifier is employed to categorize DR stages. The proposed method was validated by performing ablation studies for each pre-trained model. Specifically, classification evaluation metrics including precision, recall, F1- score, and accuracy were computed for each pre-trained model with and without the attention block. The comparison of results clearly showed the effectiveness of adding the attention blocks to each model. Further, the final MLP classifier trained on the fused convolutional features from all pre-trained models was also evaluated for DR grading, which resulted into precision, recall, F1-score, and accuracy of 0.83%, 0.81%, 0.82% and 0.90%, respectively. These results were compared with state-of-the-art methods to show that the proposed method outperforms existing methods significantly. Therefore, the proposed method can be a useful tool for providing a secondary opinion to medical practitioners for DR grading and classification.
  • No Thumbnail Available
    Item
    Strategy Prediction of Pakistan Cricket Team – Cricket Analytics
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2021) Muhammad Ahsan , Muhammad Hamza , Abdul Rafeh; SP17-BCS-003 , SP17-BCS-114 , SP17-BCS-157; Dr. Muhammad Aksam Iftikhar; LHR TP 7128
    Cricket is a famous sport and it is not possible to be sure of the outcomes of a planned strategy no matter how experienced a player is, the performance of the batsman/bowler can be affected by many factors including the environmental factors or the bowler he is facing. We can train a model with the past behaviors of a sportsman against a certain bowler on a certain pitch. This model would learn the behaviors of team players and predict the score range that player could make in that match, score range against a bowler in that match. In our project, we will provide an interface to the user using which the user can interact with the model in such a way that once the match starts the user will provide the opponent team, match location. After that, a user and model session will start in which he would be able to give live updates. e.g. the current batsman, the current bowler, if the batsman got out, on which the model would provide a live prediction on that input e.g. the chances of a batsman getting out against that bowler or the score he can score against a bowler on his best positions, the approximate score that the batsman might break in that over, the model would predict which next batsman would be suitable against that bowler and what stats he could achieve. The output that the system gives will portray the score range prediction, score range will be a bracket where the score from that batsman might fall, score range against all players, score range over wise and live match companion that interacts the way people do while watching the match.
  • No Thumbnail Available
    Item
    Segmentation of Brain Tumor using Deep Learning Techniques
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2022) Muhammad Hashir Khan; FA19-RCS-026; Dr. Muhammad Aksam Iftikhar; LHR TP 8357
    Cancer is one of the fatal types of disease in the World. A brain tumor is a type of cancer that appears in the glial cells of the brain. There exist two types of brain tumors. One is High-grade glioma and the other is low-grade glioma. The survival rate for HGG is very low while according to history most patients diagnose with LGG survive this disease. To detect the tumor in the brain one of the popular techniques is the Magnetic Imaging Resonance (MRI). It has four different modalities and neurologists after observing the different MRI modalities diagnose the tumor location and category. But manually detecting the tumor from MRI scans is a difficult task. There is always a need to segment the tumor region automatically from an MRI scan. To solve this problem, many researchers proposed different solutions. Many researchers use deep learning models to address this issue. One of the earliest networks used for this was Convolutional Neural Network (CNN) and it gives good results for tumor segmentation tasks. In this research, the approach we used is based on the ensemble method. In this approach, we use three different U-Net models to train them from scratch and predict the results on each model separately. After that, we ensemble all three models predicted results applying the majority voting technique and produce a result on testing data. Our proposed ensemble method produces a dice score of 0.86, 0.88, 0.89, and Hausdorff distance 2.0, 2.0., 2.0 for the three categories of a tumor, Enhancing tumor, Whole tumor, and Tumor core respectively on the testing dataset. Our results are better than many comparable state of the art method including the Brats 2019 challenge prominent papers.
  • No Thumbnail Available
    Item
    City Scale Multi Camera Vehicle Re-Identification
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2023) Zain Ul Abidin; SP20-RCS-030; Dr. Muhammad Aksam Iftikhar; LHR TP 8467
    Vehicle re-identification (Re-ID) is a difficult computer vision task that requires matching and tracking cars across many cameras with no overlapping views. Its applications in traffic monitoring, law enforcement, and intelligent transportation systems have piqued the interest of many people. This work addresses the issues of vehicle re-identification by merging vehicle, orientation, and camera re-identification approaches. This research aims to overcome the limits of existing vehicle re-identification systems, which struggle with comparable backgrounds, shapes, and orientations. These characteristics contribute significant bias in similarity evaluation and impede the acquisition of fine-grained information. To address these challenges and improve the accuracy of vehicle re-identification, the study offers VOC-ReID, an orientation re-identification approach. Furthermore, VGG-NeT with Global Average Pooling is used to extract more distinguishing characteristics from vehicle photos. The triplet of vehicle-orientation-camera is considered a unitary entity by VOC-ReID, changing the focus from backdrop and form to re-identification. Numerous experiments were carried out utilizing large-scale vehicle re-identification datasets, including real-world surveillance film, to assess the effectiveness of the VOC-ReID technique. In comparison to our baseline technique, we observed a slight improvement in our results. In our approach, we divided the entire image into 360 bins and determined the direction of each bin. These directions were utilized as orientation labels in the cityFlow dataset. When we applied our technique to the cityFlow dataset, we achieved an accuracy of 67% in terms of mean Average Precision (mAP). The accuracy of the method was assessed using mean Average Precision (mAP) and Rank-1 accuracy measures.
  • No Thumbnail Available
    Item
    An Automated Diabetic Retinopathy Grading Pipeline based on Deep Learning Approaches
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2023) MUHAMMAD ZAHID AHMAD; FA21-RCS-007; LHR TP 8471; Dr. Muhammad Aksam Iftikhar
    Early detection is essential for efficient disease treatment and prevention since diabetic retinopathy (DR) is one of the main causes of blindness in the globe. This thesis addresses the value of early detection during the DR stages and the part deep learning plays in creating a reliable detection system. The study stresses the value of early DR stage detection to stop disease development and enhance patient outcomes. Early illness detection and classification allow for the implementation of prompt therapies that can halt or delay the disease's course. In addition to increasing the likelihood that visual function will be preserved, this lessens the need for expensive and intrusive therapies, which eases the strain on healthcare systems. To achieve accurate detection of DR, a deep learning approach is employed. Specifically, an ensemble method of two Vision Transformer (VIT) models followed by Support Vector Machine (SVM) is utilized. We have used Kaggle APTOS 2019 dataset. The ensemble model capitalizes on the strengths of multiple VIT models to enhance feature representation and improve classification performance.The experimental results demonstrate the effectiveness of the proposed model, achieving an overall accuracy of 91.14%, precision of 92%, recall of 91%, and an F1 score of 91% on APTOS dataset. These results outperform existing methods and validate the potential of deep learning in DR detection. The developed model showcases the capability of deep learning algorithms to provide reliable and efficient diagnostic support, enabling early intervention and ultimately contributing to the prevention of DR-related visual impairment.
  • No Thumbnail Available
    Item
    Survival Prediction of Glioma Patients Based on Radiomic Features
    (Library Information Services, COMSATS University Islamabad, Lahore Campus, 2023) Talha Naeem; FA19-RCS-029; LHR TP 8464; Dr. Muhammad Aksam Iftikhar
    Our brain is the most important part of our body that controls and commands all the functions. It Takes information from different organs of the body and five main senses and then commands the tissues to perform specific tasks. Brain administrates the motions, creativity, emotions, and memory. The brain is enclosed within the skull, and it consists of Cerebrum, Cerebellum and Brainstem. Cerebrum hemisphere portion is further split into lobes. Each hemisphere includes the Temporal Lobe, Frontal Lobe, Parietal Lobe, and Occipital Lobe. All these hemispheres are once further divided into their areas to accomplish the specific factions. Each lobe cannot accomplish any of the body function without the help of other lobes separately. The relationship between the lobes of the brain and left and right hemisphere is very complex. The left hemisphere deals and commands to the right portion of body and right hemisphere deals and commands to the left portion of body. Gliomas are unique and very extensively recognized categories of the brain tumor. Glioma is a brain tumor category that starts producing in the brain or the spine glial cabals. It is a kind of sturdy cell in our brain. For precise and strong diagnosis of glioma patients both the segmentation into sub-regions and overall survival (OS) prediction tasks are very important. These tasks are used for the better treatment arrangement and risk influence investigation. Form the past couple of year’s investigation in brain tumor has seen remarkable progress. Many researchers present admirable work for better treatment of glioma patients, better treatment arrangement and OS prediction time. However, still to the current era, huge gap exists for precise segmentation and OS prediction task of the glioma patients. Researchers are trying to progress in the complexities and challenges of brain tumor. In this proposed methodology, our set of rules achieve good performance in OS prediction task of glioma patients by using machine learning outline. We perform segmentation of brain sub-regions on magnetic resonance imaging (MRI) scans. Based on segmented sub- regions, we extract the radiomic features such as intensity, shape, and texture. The proposed approach employed on multimodal brain x tumor segmentation (BraTS) with OS prediction task for BraTS-2020 dataset and trained on the random forest (RF) model for the OS task for the prediction of survival days for brain tumor patients. The evaluation of the performance measures is carried out for segmentation process by a pre-processing technique and OS prediction by measuring accuracy of the RF model and we obtained the results of 59% and 51% for training and testing of glioma patients respectively.