My research focused on the deployment and identification of novel implementations of AI in real-world healthcare that have a positive
impact on patient outcomes and provide clinicians with a decision-support system to improve clinical practices.
Projects
I have a passion for developing ML models in areas related to regression Analysis, single and multilabel classification, image processing, image classification and extraction, natural language processing, and digital signal processing.
- EEG data driven Machine Learning for classification of stroke and healthy subjects: Utilized python to pre-process EEG data of 21 subjects total of 1000 trials, implemented 14 supervised ML algorithms (e.g., DNN, KNN, Decision tree, LightGBM, RF, Extra-trees Classifier) to classify stroke patients with 96.5% accuracy, and detected the regions of neural activity using explainable AI.
- Classification of stroke patient from gait cycle using traditional ML algorithm: Preprocessed unstructured image data of unique gait cycles of 10 subjects and performed exploratory analysis, applied 6different image registration techniques to classify diabetic neuropathy patients with 98% accuracy.
- Brain MRI segmentation: UNet architecture to segment brain tumors from MRI images and performed an analysis of 110 patients’ registered images and corresponding masks, the model achieved an average of 92% IoU on test data.
- Automatic Breast Tumor Segmentation and Classification: A Deep Learning Approach: Developed a CAD system that can accurately segment regions of interest in ultrasound images of the breast and classify tumors as benign, normal, or malignant. I used the Breast Ultrasound Dataset to train and test the model and utilized the U-Net architecture, a deep-learning framework that has shown excellent performance in medical image segmentation tasks.
- Image segmentation from CT Scan: Developed an image processing scheme with a graphic user interface model to segment and quantify lung tumors and emphysema using lung CT images
Journals
Apriori-backed Fuzzy Unification and Statistical Inference in Feature Reduction: An Application in Prognosis of Autism.
S. Maitra, N. Akter, A. Mithila, T. Hossain, & M.S. Alam. (2020).
In 5th International Conference on Advanced Computing and Intelligent Engineering.
Applied Apriori algorithm and novel preprocessing techniques to optimize Neural Network (NN) performance. Modeled huddling features using fuzzy techniques, then eliminated features using statistical tests. NN achieved 99.68% accuracy with reduced computational cost.
Prediction of Academic Performance Applying NNs: A Focus on Statistical Feature-shedding and Lifestyle.
S. Maitra, S. Eshrak, M.A. Bari, A. Al-Sakin, R.S. Munia, N. Akter, & Z. Haque. (2019).
International Journal of Advanced Computer Science and Applications, 10(9).
Implemented a machine learning algorithm to analyze the multiple factors contributing to academic excellence, including psychology, habits, lifestyle, and preferences.
Master’s Project
Detection of Autism Spectrum Disorder applying Deep Neural Network
Developed a DNN classifier for early ASD diagnosis, improving symptoms and function in toddlers. Fine-tuned model with diverse data and considered ethical considerations for transparent and unbiased predictions.
Bachelor Thesis
Acoustic Echo Cancellation for the advancement in Telecommunication.
Designed and implemented an advanced echo cancellation system using the Frequency Domain Adaptive Filter (FDAF) method, resulting in improved performance. Analyzed frequency components and adapted filter coefficients to effectively remove echo in complex environments.