School of Engineering Department of Computer Science and Engineering 91 AI in Medical Imaging: Automatic Stroke Analysis on Brain CT Scans Supervisor: CHEN Hao / CSE Student: SON Moo Hyun / DSCT Course: UROP 2100, Fall Ischemic Stroke, a serious disease with a high mortality rate, is typically diagnosed using CT scan derived perfusion parameter maps. Patients in their acute stage have symptoms not limited to Ischemic Stroke but can be symptoms of another fatal disease such as brain tumor. The perfusion parameters maps are used by the doctors to diagnose whether the symptom is due to stroke. Here, the need for an accurate perfusion parameter map arises. Currently, the perfusion parameter maps are mathematically derived using RAPID software, which requires explicitly choosing an AIF by Radiologist. In this paper, we propose a model capable of producing AIF-free perfusion parameter maps. Besides, patients are exposed to a high radiation dose during the imaging stage. Various attempts have been made to address this: reducing the bolus injection or lowering the temporal resolution. However, these efforts have not yielded significant results. In this paper, we propose a deep neural network model capable of producing perfusion parameter maps with up to 1/16th of the full-temporal resolution indistinguishable from that of fulltemporal resolution RAPID. Data-efficient, Domain Generalizable and Interpretable Deep Learning Supervisor: CHEN Hao / CSE Student: HUANG Zhihua / COMP Course: UROP 1100, Fall During the past three months from September to November of this semester, I joined the Ophthalmology Group, under the guidance of Ph.D. candidate CHE Haoxuan. I enriched my knowledge of deep learning by studying the material about it like Dive into Deep Learning and the hands-on experience in doing a project focused on the data pre-processing in Federated Learning by applying the Dirichlet distribution. Furthermore, I actively joined the group meetings where different undergraduate students in the same group share their ideas about respective topics on their projects and papers in different conferences like CVPR, and ECCV, which greatly broadened my horizons in the area of computer vision and deep learning. Data-efficient, Domain Generalizable and Interpretable Deep Learning Supervisor: CHEN Hao / CSE Student: TAN Juin / COMP Course: UROP 1100, Spring Breast cancer remains a significant global health concern, causing an estimated 670,000 deaths worldwide in 2022 (WHO, 2024). Despite advances in medical technology, identifying cancer cells manually is challenging due to subjectivity and interpretational variability. To address this issue, machine learning techniques, specifically deep learning-based pipelines, have been proposed as a potential solution for assisting in breast cancer diagnosis through the analysis of FISH (Fluorescence in-situ Hybridization) medical images. The report details the flow of the investigation and identifies the experiment's challenges. Moreover, the report emphasizes the importance of health promotion for early detection, timely diagnosis, and comprehensive breast cancer management as the three pillars towards reducing global breast cancer mortality. By reducing global breast cancer mortality by 2.5% annually, 25% of breast cancer deaths could be averted by 2030 and 40% by 2040 among women under 70 years of age (WHO, 2024).
RkJQdWJsaXNoZXIy NDk5Njg=