Md. Maruf - Research

Deep Learning Based Acute Lymphoblastic Leukemia Detection from Peripheral Blood Smear

In my Undergraduate thesis, I developed a deep learning-based system for detecting Acute Lymphoblastic Leukemia (ALL) using peripheral blood smear images. The research involved training various pre-trained models, with and without custom architectures, to enhance diagnostic accuracy and efficiency. This automated system offers faster and more precise detection compared to traditional manual methods, significantly reducing the time and errors involved in diagnosis. By leveraging deep learning, the solution can be scaled for other hematological disorders, ensuring adaptability and robustness. This innovation paves the way for improving clinical workflows, leading to quicker treatment decisions and better patient outcomes. Figure 1 shows the general flowchart of ALL detection using D-CNN model used in my research.

Supervisor : Md. Mehedi Hasan
Website: RUET-Rajshahi University of Engineering & Technology.

Figure 1: General flowchart of ALL detection using D-CNN model

Figure 2: Step by step image preprocessing

Figure 3: Performance of custom architecture on test dataset

Intelligent Fire Surveillance: A Deep Learning Framework for Accurate Indoor Fire Detection

Motivated by a severe fire disaster in Bangladesh, I developed a cost-effective fire detection system using deep learning. My model, 'Intelligent Fire Surveillance: A Deep Learning Framework for Accurate Indoor Fire Detection,' provides early detection of indoor fires with high precision. This innovative solution is designed to enhance safety in public and private spaces, offering a reliable, AI-driven alternative to traditional fire alarms. The framework leverages advanced neural networks to detect fire hazards swiftly, making it a critical tool for preventing fire-related tragedies.

Doi: https://doi.org/10.1109/RAAICON64172.2024.10928378

Figure 4: Flowchart of the Methodology

Figure 5: Images after preprocessing

Figure 6: Confusion Matrix

Figure 7: Number of images in each class.

State-of-the-Art Review on Fall Prediction among Older Adults: Exploring Edge Devices as a Promising Approach for the Future

Falling is one of the most severe threats to the health and well-being of older people, which has a consequence on their daily activities and standard of living. Additionally, the cost of treating injuries brought on by falls is high, and in some situations, the patient may not make a full recovery. Existing fall prediction approaches aim primarily on biological characteristics in particular locomotion, eyesight, and cognition, ignoring the fact that falls are complex by nature. In this paper, we comprehensively reviewed state-of-the-art fall prediction systems and listed out different factors directly associated with falls. We analyzed the current trends and extracted that machine learning, deep learning, sensors, and gait-based fall prediction methods are some of the most prevalent technologies. However, TinyML-based smart wearable technology and gadgets have the real potential to predict intricate physiological phenomena like falls. As a result, this research advocates the emergence of TinyML-based smart wearable gadgets to support geriatric people in preventing falls This research is likely to significantly boost the quality of life for the geriatric population by giving future researchers a greater understanding of the systems that are now available.

Doi: https://doi.org/10.1016/j.measen.2025.101878

Supervisor : Md. Mehedi Hasan
Website: RUET-Rajshahi University of Engineering & Technology

Figure 8: The figure shows fall prediction methods for acquiring data from a subject and sending the data to the cloud data center as well as a basic machine learning process.

Figure 9: The TinyML-based fall prediction system framework

Figure 10: Roadmap to future fall prediction systems from current technologies