← Back to Explore

Novel Convolutional Neural Networks for Improved Accuracy in User-Accessible Brain Tumor Detection and Classification

JSHS · 2024

Overview

Effective treatment for brain cancer is aided significantly by the rapid detection of tumors. Traditional detection methods involving the manual inspection of Magnetic Resonance Imaging (MRI) scans have been established to be slow, consuming large amounts of time both for the specialist and the patient. The purpose of this design investigation was to propose and develop three artificial intelligence (AI) models, specifically, convolutional neural networks (CNNs), employing novel techniques to quickly and ac curately detect the presence and type of brain tumors from MRI scans. The goal was to provide medical professionals with a faster alternative compared to traditional, manual detection methods by overcoming inefficiencies and reducing human error. The design was also able to address various gaps and limitations in existing studies, most notably, the lack of a user interface (UI) for easy and practical accessibility. Therefore, in this design, the highest performing CNN, utilizing the pretrained VGG19 archite cture—with a validation accuracy of 98.06% and validation loss of 0.02 —was integrated into a minimalistic, open -source UI website using the Gradio library, enabling hospital workers to upload MRI images with ease for tumor identification, facilitating real-world applications and usage in various understaffed medical settings. Algorithmic Design and Computational Modeling Using Dynamic Spectrum Allocation Techniques to Optimize Bandwidth Management in Wireless Communication Systems Ankit Walishetti Illinois Math and Science Academy, Aurora, IL Mentors: Dr. Randall Berry and Dr. Igor Kadota, Northwestern University This study aims to address the pressing need for efficient spectrum management methodologies in wireless communication systems by developing innovative sorting and allocation algorithms. Leveraging Dynamic Spectrum Allocation (DSA) techniques, this resear ch seeks to devise strategies to optimize the utilization of bandwidth within existing spectrum space, ultimately reducing the need for spending and network infrastructure expansion. Ensuring thorough coverage of DSA techniques, 5 distinct transmitter sort ing algorithms were programmed and tested across 8 performance metrics designed to measure specific capabilities. For consistency, a single bandwidth allocation program was designed to ‘pack’ transmitters starting from the left endpoint of the spectrum spa ce. Progressively varying the transmitter count, Northwestern’s Quest Supercomputer performed the final computation, using 64 gigabytes of RAM and running for ~10 hours. The sorting algorithms based on the product of radius and bandwidth (Power Sort) and based on the most signal interference (Most-Overlap Sort) were most efficient, performing well in two key categories. Most-Overlap Sort and Power Sort produced average feasibility values of 0.675 and 0.664, respectively, scoring a 8.6% – 10.6% improvement i n feasibility performance. Consequently, the greater maximum bandwidth remaining indicated highly efficient allocation within spectrum resources; Power Sort and Most-Overlap Sort took up 8.888 and 9.042 bandwidth intervals, respectively, performing ~6% bet ter than control. Most-Overlap Sort allocates the most ‘problematic’ transmitters first, gaining an edge in end feasibility. Power Sort’s success can be attributed to its prioritization of bandwidth/area-hungry transmitters, simplifying the allocation of minor transmitters in the final stages. Indiana Identifying Pyroptosis-Related Genes in Alzheimer’s Disease Based on Bioinformatics Analysis Divya Ariyur Carmel High School, Carmel, IN Alzheimer's disease (AD) is the seventh leading cause of death in the United States, with over 6 million Americans reported in 2023. Recent studies have shown the association between AD and inflammation. Pyroptosis, a newly discovered form of cell death, i s associated with inflammation and has been found to be involved in AD. However, the exact relationship between pyroptosis and the pathology of AD remains unclear. The purpose of this project was to identify the pyroptosis -related genes (PRGs) associated w ith AD and to analyze their roles in the disease pathology. To this end, microarray data analysis was performed on the GSE48350 dataset containing the post -mortem brain tissue samples from 80 AD patients and 173 healthy individuals. The differentially expr essed genes that intersected with 52 PRGs from published literature were noted as differentially expressed pyroptosis-related genes (DEPRGs), which included 4 up- regulated and 4 down-regulated genes. Gene Ontology and Kyoto Encyclopedia of Genes and Genome s enrichment analyses revealed that the DEPRGs were enriched in the positive regulation of cysteine -type endopeptidase activity involved in the apoptotic process, cAMP -dependent protein kinase complex, cysteine-type endopeptidase activity involved in apoptotic signaling pathway, and Legionellosis. Four hub genes, SCAF11, CASP8, CYCS, and TP53, identified using Cytoscape software, were significantly expressed in AD patients, and their diagnostic and predictive value was evaluated using receiver operating characteristic curves. These findings suggest that DEPRGs play a crucial role in the development and progression of AD and could be used as potential diagnostic biomarkers and therapeutic targets for AD treatment. Quantum Annealing for the Set Splitting Problem Sean Borneman Bloomington High School South, Bloomington, IN This work explores a novel application of quantum annealing to solve the Set Splitting Problem and common variants, including Max Set Splitting and Weighted Set Splitting. The Set Splitting Problem is applicable to DNA micro -array data analysis, graph base d cybersecurity, and other fields. I propose a quadratic unconstrained binary optimization (QUBO) problem formulation of the Set Splitting Problem in order to encode any given problem onto a Quantum Annealing Computer. The key contribution of the work consists in formulating penalty functions that ensure the ground state of the QUBO Hamiltonian corresponds to valid solutions that split the input subsets. This approach scales linearly in terms of the number of logical qubits needed relative to problem size. Empirical tests of the proposed solution show convergence to globally optimal solutions, with high accuracy rates over repeated trials. One limitation of the proposed solution is in handling sets with cardinality above 3, in that the achieved solution migh t not be optimal; however, solution optimality can be checked. Hardware limitations of current quantum annealers lead to an exponential rise in required physical qubits, versus the theoretical linear increase, although this can improve with future developm ents. Further work is needed to enhance formulation robustness, reduce qubit requirements for embedded problems, and to conduct more extensive benchmarking. Quantum solutions to the Set-Splitting problem lead to reduced time complexity versus classical solutions, and may accelerate research in biology, cybersecurity, and other domains.

Competition history

  • JSHS 2024 Category not listed

Resources

Related projects

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

Source: Junior Science and Humanities Symposium

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google