Predicting a Diagnosis, Prognosis, and Treatments for Neurodegenerative and Cancerous Diseases
Overview
According to the WHO, neurodegenerative diseases and cancer are progressively occupying a larger portion of mortality across the world, constituting 10 million new cases and over 600,000 deaths per year in the United States alone. Despite early diagnosis and estimating the future of course of terminal illnesses being integral for survival of patients, the healthcare gap and limited access to hospitals, facilitated by the COVID-19 pandemic, has made it difficult to access accurate screenings for these diseases. Currently, clinical and algorithmic methods fail to effectively utilize the multimodal data available, are time-inefficient, expensive, and inaccurate. The objective of the research is to create a novel end-to-end quantum machine learning approach using multiple data modalities for the prediction of a diagnosis, prognosis, and identification of effective treatments. In a procedural flow, data can be collected from one or more of the following: CT scan images, webcam, patient-physician audio, Whole Slide Images, and clinical data. After preprocessing, to ascertain a diagnosis, a pretraining process and specialized model architectures are developed for each data type. For image data, a Quantum Convolutional Neural Network (QCNN), is employed to detect high level features. With text-based clinical data (including the audio-derived data), a Bidirectional Encoder Representation (BERT) model is used to extract text embeddings. For video data, a python library, OpenCV, is used to craft pupil progression and average fixation duration features. All feature vectors are concatenated, normalized, and passed through a quantum node, optimized by Stochastic Gradient Descent, and then mapped to one of 38 neurodegenerative and cancerous diseases. For prognosis, the quantum features are pooled, concatenated with the diagnosis feature vector and passed through a gradient boosting neural network with an output of survival times. Treatment prediction is an information retrieval task, based on a knowledge graph linking research papers related to drug/treatments approved by the Federal Drug Administration. The proposed approach was tested on 5,000 patient profiles sourced from publicly available databases and was shown to accurately predict diagnoses with an F-Score of 95.70% and an accuracy of 98.53%. For predicting prognoses, the model achieved a C-index (Concordance Index) of 0.94, outperforming previous state-of-the-art approaches by over 15%. Due to the need for manual review, treatment prediction was tested on only 500 patient profiles, achieving a 99.32% accuracy and a 97.42% F-Score. OmniDoc is significantly cheaper, faster, and more accessible than contemporary methods, facilitating life-saving diagnostics, prognostics, and treatment prediction in areas where in-person healthcare services are limited.
From the student
Computer Science always fascinated me in elementary school. The idea that you could program a computer to do exactly what you would tell it to do, allowing for the automation of several tedious tasks baffled me. I continued to learn and expand my knowledge of this mystifying field, until I had developed several apps that had attained thousands of downloads. I remember in 7th grade, I was introduced to the concept of Machine Learning, which shattered my view of computer science. I was shocked by how a computer could learn to do things on its own, and even more, how it was being applied in several intensive real-world scenarios.
I knew then and there I wanted to delve into Machine Learning, but I didn't want to wait until I was in my 20s to begin researching the topic. I also aspired to make a positive, lasting impact in my community. This became even more apparent in 7th grade, when my uncle was diagnosed with terminal cancer, despite consistent screenings with doctors. As such, I conducted my first research project in 8th grade regarding the diagnosis of lung cancer, the disease my uncle suffered from, with neural networks. This project won several awards, including 1st place at my regional science fair, semifinalist in the national Broadcom MASTERS competition, and a medal at the state science fair. In the summer before my freshman year, I interned at a healthcare startup, where I learned to use several novel machine learning models, specifically pertaining to natural language processing. This year, in 10th grade, I developed a universal framework that utilizes several novel machine learning models in the areas of audio analysis, video analysis, image analysis, and natural language processing to predict diagnoses, prognoses, and treatments for a given patient. I received 1st prize at the Texas Junior Academy of Sciences (TJAS), for this project and am presenting this project at the 2022 American Association for the Advancement of Science Annual Meeting. I also received 1st prize and 2nd grand prize at the Texas Junior Science and Humanities Symposium (TJSHS) for this project, additionally being invited to present my research at the National Junior Science and Humanities Symposium (NJSHS).
Images (19)
Awards (1)
- AJAS Fellows Badge
Competition history
- AJAS 2022
Resources
Related projects
JSHS · 2022
OmniDoc: A Multimodal Quantum Machine Learning Approach to Diagnosis, Prognosis, and Treatment Prediction for Neurodegenerative and Cancerous Diseases
CSEF · 2014
Indium: Using Novel Machine Learning Algorithms to Develop a Nondisease-specific Personalized Medicine Engine
ISEF · 2022
A Quantum Machine Learning-Based Framework for Early Cancer Detection and Biomarker Identification Through Transcriptome Profiles
JSHS · 2025
QLC-Net: A Hybrid Quantum Support Vector Machine infused Convolutional Neural Network Approach to Detect Bronchogenic Carcinoma (Lung Cancer) in Pathological Slides
ISEF · 2026
JomedAI: A Fully Deployed Artificial Intelligence Platform Integrating 23 Independent Deep Learning Models Across 13 Organ Systems, 1,808 Validated Clinical Statistical Scoring Models, and AI-Powered Drug-Drug Interaction Detection for Early Disease Diagnosis, Interactive Clinical Reasoning, and Medication Safety
AJAS · 2026
Agentic AI Cancer Treatment Plan Generation
ISEF · 2023
Implementing a Hybrid Quantum-Classical Neural Network by Utilizing a Variational Quantum Circuit for Detection of Dementia
ISEF · 2014
Indium: Using Novel Machine Learning Algorithms to Develop a Nondisease-specific Personalized Medicine Engine
Closest projects by meaning, across every fair and year in the corpus.