Muhammad Zain Amin
| Home |
|---|
| Projects |
| Publications |
| Awards |
| Services |
| Resume |
Projects
Two-Step Hierarchical Classification of Skin Lesions Using Transfer Learning and Machine Learning Techniques
This Project was implemented using ISIC2017 dataset, to classify the skin lesion categories such as benign, melanoma, and seborrheic keratosis. A hybrid model, with a combination of Xception and Random Forest, achieved the maximum Balanced Multiclass Accuracy (BMA) score of 79%, outperforming the VGG16, InceptionResNetv2, and DenseNet201 architectures
Breast Mass Detection and Segmentation Using Multi-scale Morphological Sifting and K-Means Clustering
This project was implemented for mammographic mass detection and segmentation using a multi-scale morphological sifting approach integrated with a mean shift filter, k-means, and post-processing that detects and segments masses. This implementation was evaluated on InBreast dataset and was able to segment mass lesions with a sensitivity of 76.92%.
Alzheimer Disease Classification Using MRIs and Gene Expression Data
This project was implemented for developing three binary classifiers with various feature selection and validation methods for classifying between Alzheimer disease macro-stages, which are CTL, MCI, and AD. The dataset used contained both MRI and Gene expression features for various number of patients. The final result was evaluated using AUC and MCC metrics.
Regularized General Eigenvalue Classifier (ReGEC)
The project was aimed to implement ReGEC algorithm from scratch using R-programming language using different kernel types, such as linear and gaussian. The performance was evaluated using 4 datasets, which are Cleveland Heart Disease, Pima Indians, Breast Cancer, and German Datasets. The overall performance was significantly close or higher than the results reported in the paper.
A Comparative Study on the Automated Detection and Diagnosis of Diabetic Retinopathy
This project focused on comparing methods for automating the detection and diagnosis of Diabetic Retinopathy (DR), aiding eye care professionals. Utilizing classification and segmentation techniques, particularly in retinal fundus image assessment, has shown promising results for early DR diagnosis. The project extensively analyzed recent cutting-edge approaches for DR classification and segmentation, encompassing various techniques including machine learning, deep learning, ensemble learning, and attention mechanisms.
Detection of Parkinson’s Disease Using Automated Tunable Q Wavelet Transform Technique with EEG Signals
This project employed an automated Tunable Q Wavelet transform (ATQWT) technique on EEG signals to detect Parkinson's disease. EEG signals accurately capture brain changes in Parkinson's disease. The signals were divided into sub-bands (SBs) using ATQWT to extract detailed information. This method automatically chose tuning parameters, enhancing SB representation and signal reconstruction. The approach achieved high accuracies of 96.13% and 97.65%, with area under the curve values of 97% and 98.56% for distinguishing between healthy controls and Parkinson's patients in both medication states (OFF and ON), respectively, using the least square support vector machine.
Comparative Analysis of Eight Direction Sobel Edge Detection Algorithm for Brain Tumor MRI Images
This project employed the eight-directional Sobel Edge Detection algorithm to identify brain tumors in MRI images. The goal was to precisely detect shapeless tumor growth using improved segmentation techniques. The study introduced a Sobel algorithm with an eight-directional template to enhance edge detection in MRI brain scans. This approach was compared to traditional methods, evaluating its performance through metrics like MSE, RMSE, Entropy, SNR, and PSNR.