Imagine one day, your breathing became consistently labored and shallow. Months later you were finally diagnosed with pulmonary fibrosis, a progressive disease that naturally gets worse over time with no known cause and no known cure, created by scarring of the lungs. If that happened to you, you would want to know your prognosis. That’s where a troubling disease becomes frightening for the patient. Outcomes can range from long-term stability to rapid deterioration, Natural history of IPF is unknown and the prediction of disease progression at the time of diagnosis is notoriously difficult and doctors aren’t easily able to tell where an individual may fall on that spectrum. Data science, may be able to aid in this prediction. If successful, patients and their families would better understand their prognosis when they are first diagnosed with this incurable lung disease. Improved severity detection would also positively impact treatment trial design and accelerate the clinical development of novel treatments.

Lung function is assessed based on output from a spirometer, which measures the forced vital capacity (FVC), i.e. the volume of air exhaled. Our aim is to predict a patient’s severity of decline in lung function based on a CT scan of their lungs, metadata, and baseline FVC as input. We want to predict the final three FVC measurements for each patient, as well as a confidence value in our prediction.

In our work we have shown how we can use deep learning technique to predict the idiopathic pulmonary fibrosis progression. This type of prediction analysis, if applied effectively in medical sectors, can help patients by analysing their lungs condition from Ct scan and the corresponding other information so that treatment can be started as early as possible. Early treatment can improve the survival rate of a patient. Thus,deep learning can help medical practitioners to understand their prognosis in a better way when they are first diagnosed with idiopathic pulmonary fibrosis. Hence, deep learning algorithms are adding significant value to the healthcare industry as well as to the society aiming towards a healthy and normal social lifestyle.

Over the last few decades, enormous researches have been done on WSO which has mainly emphasized the physical model development rather than the generation of the data model. However, the data model is crucial for numerical analysis since it reflects all the characteristics of the physical model and easy to handle for optimization. Due to the presence of redundant data, the smooth computation can never be performed on the unoptimized mathematical model. Therefore, ROM techniques can be applicable to turn the unoptimized data model optimized by reducing the useless dimension from the data model. As a result, the physical model generated from the optimized data model becomes smooth to analyze and gives an optimal geometric shape easily. Dealing with WSO using ROM techniques is a new concept that is excepted to complete flawlessly in this project.

In this research work, we have developed mathematical algorithms and software for the model reduction of large-scale dynamical systems. The proposed algorithms and software will be used for academic research and industrial applications. To show the efficiency and capability of these algorithms we have applied them to a set of real-world data obtained from a research institute and generated by ourselves. The results obtained in this work were presented at national and international conferences. We also have published several Journal and conference papers using the models and numerical results obtained from this project. Besides, two M.Sc. thesis works have been completed by the funding of this project.

Publication from this research:

1. Md. Tanzim Hossain, Azizul Haque, Kife I. Bin Iqbal, and M. Monir Uddin, Airfoil Generation using GANs and Optimization using Model Order Reduction. (In preparation)
2. Kife I. Bin Iqbal, Md. Tanzim Hossain, and M. Monir Uddin, Generation of the Mathematical Model to Analyze the Dynamical Behavior of the Turbulence Flow Around the Airfoils. (In preparation)
3. A. Haque, T. Hossain, M. Murshed and K. I. B. Iqbal and M. M. Uddin, Estimating aerodynamic data via supervised learning, 25th International Conference on Computer and Information Technology (ICCIT), IEEE, Dhaka, 2020.

Over the last few decades, enormous research has been done on STSO, which has mainly emphasized the physical model development rather than the data model generation. However, the data model is crucial for numerical analysis since it reflects all the characteristics of the physical model and is easy to handle for optimization. Due to redundant data, the smooth computation can never be performed on the unoptimized mathematical model. Therefore, ROM techniques can be applicable to turn the unoptimized data model optimized by reducing the useless dimension of the data model. As a result, the physical model generated from the optimized data model becomes smooth to analyze. Dealing with STSO using ROM techniques is a new concept that is excepted to complete flawlessly in this project. The possible significant outcomes of this research are as follows:

• The physical model of the PV panel will be constructed, and data on the solar radiation on the PV panel will be extracted to generate a mathematical model.
• By linearization procedure, the linear state-space formation will be generated based on the input-output relations, which will mainly define the properties of the physical model.
• The generated mathematical models' ROM will be created to truncate the redundant design meshes to boost up the entire simulation procedures.
• Hopefully, the optimized PV panel model will be found in this project to improve the physical efficiency in terms of thermal state and energy storage capacity and minimize the loss of energy due to heat.
• The project will help acquire knowledge on STSO in terms of Control theory, Scientific computing, and Mathematical Algorithms.
• Several scientific papers will be published in renowned journals highlighting the outcomes of this project.

Multiple Sclerosis (MS) is a chronic inflammation of the central nervous system that causes the destruction of the protective sheath of nerve cells, resulting in degraded masses known as plaques. This cellular damage to the brain or spinal cord can lead to various side effects such as impaired vision, loss of balance, and learning disorders.

Currently, the diagnosis and progress of MS are determined visually by comparing images of the brain at different periods. This process is time-consuming and challenging for specialists due to the small size of the lesions, the number of lesions, and their spatial distribution, as well as the different severities of lesion progression (white, gray, and black holes). Diagnosis of MS involves examining the clinical symptoms of patients suspected of having MS and using MRI images of the brain. These images show the location and number of lesions in the white matter area of the brain, which are essential for diagnosis and follow-up.

However, accurately diagnosing MS-related lesions can be challenging because various diseases can cause lesions in the brain. The current standard of manually segmenting images to specify the size and location of lesions is time-consuming and less accurate than computer diagnostics, leading to non-identical diagnoses by professionals. Moreover, confuent lesions, which overlap and cannot be easily separated, further complicate manual segmentation.

To address these challenges, researchers have developed a deep learning-based model using the U-Net deep neural network, in which the wavelet pooling layer replaces the max pooling layer. Wavelets allow for localization in scale (i.e., frequency) and space, which can be used to analyze local, spatial transients in the data such as edges or surfaces. This model performs better segmentation of lesions of different sizes by enabling the reduction of image size and the ability to highlight the specificity of MS lesions in MRI images. The use of deep learning in medical image information analysis, such as noise reduction, segmentation, and classification, has had good results in recent years.

Prostate cancer segmentation and detection involve the use of medical imaging techniques and computer algorithms to identify and locate cancerous regions within the prostate gland. The process can help physicians to accurately diagnose and stage prostate cancer, which can help guide treatment decisions.

The main steps involved in prostate cancer segmentation and detection include image pre-processing, image segmentation, feature extraction, and classification. Image pre-processing involves cleaning up the medical images to improve the accuracy of the segmentation and detection algorithms. Image segmentation involves identifying and isolating the regions of interest within the medical images. Feature extraction involves extracting relevant features from these regions, such as texture, shape, and size, which can be used to differentiate between cancerous and non-cancerous tissue. Classification involves using machine learning algorithms to classify the regions of interest as cancerous or non-cancerous.

Several medical imaging techniques are commonly used for prostate cancer segmentation and detection, including Magnetic Resonance Imaging (MRI), Computed Tomography (CT), and ultrasound. Deep learning techniques such as convolutional neural networks (CNNs) have also been used to improve the accuracy of prostate cancer detection and segmentation.

Prostate cancer segmentation and detection can help physicians to make informed decisions regarding treatment options, including surgery, radiation therapy, and chemotherapy. It can also help monitor the progression of prostate cancer over time and identify any changes in the cancerous regions that may require additional treatment or monitoring. Overall, prostate cancer segmentation and detection can improve the accuracy of prostate cancer diagnosis and staging, leading to improved patient outcomes.