California Southern EMBER: A Novel Quantum Computing Framework for Early Diagnosis and Predictive Biomarker Identification of Lung Cancer

Overview

Lung cancer is the most deadly form of cancer, responsible for over 2 million deaths annually. Its fatality is largely a result of difficulty in early detection, with survival rates dropping 57% when detected late. The standard for lung cancer diagnoses has been CT scans, yet they lack efficiency and accessibility, resulting in over 75% of lung cancer cases going undetected until late stages. Recently, researchers have used Machine and Deep Learning to improve upon traditional approaches, but these computational methods still fall short due to issues such as overfitting and feature isolation. In order to improve upon current computational approaches to lung cancer diagnosis, EMBER leverages quantum computing, effectively avoiding the limitations of traditional computational models. Using gene expression data collected from inexpensive and acces sible blood tests, EMBER predicts the presence of lung cancer as well as its expected progression, allowing doctors to make rapid decisions regarding patient treatment. EMBER utilized fundamental quantum properties such as superposition and entanglement to analyze gene expression in a more biologically applicable manner. In doing so, EMBER achieved an accuracy of 93% for lung cancer diagnoses on a large, diverse patient dataset as opposed to the 85% accuracy of the baseline classical Deep Learning model. In addition, EMBER identified a novel biomarker of lung cancer in microRNA gene 4456. Overall, EMBER allows for accurate and affordable lung cancer diagnoses, substantially increasing early detection and, thus, survival rates for lung cancer patients of all backgrounds, demographics, and socioeconomic statuses. Attention Based Tracking Head for Multiple Object Tracking in Autonomous Vehicle Perception Systems Lukas Cao Ruben S. Ayala High School, Chino Hills, CA Autonomous vehicles have a high potential for safety benefits and are still being developed. Many perception systems in autonomous driving are equipped with cameras to develop an understanding of the driving environment. Since information about the movemen t of pedestrians and other cars over time is crucial for handling possible changes to the driving situation, perception systems are faced with the Multiple Object Tracking (MOT) problem. Solving the MOT problem requires temporal information from prior image frames of the driving scene to be exploited. However, the optimal method for incorporating temporal information into object detection, localization, and tracking remains unclear. In this paper, we introduce an attention-based tracking head, which incorpora tes temporal information exclusively into the tracker head. By employing an experimental design paradigm, we trained and evaluated our MOT model using a subset of the BDD (Berkeley Deep Drive) 100K dataset. After comparing our model performance with the QDTrack baseline MOT model through quantitative analysis, we find that our proposed model architecture design is a plausible solution to tracking. In the context of the autonomous driving literature, this distinct approach to tracking objects via an attention mechanism for the perception system clarifies the contribution of our work.

Competition history

• JSHS 2024 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.