- Title Information
- Title
- A Deep Learning Method Predicting Lung Adenocarcinoma Recurrence: Binary and Time-Based Approaches
- Type of Resource (primo)
- dissertations
- Name:
Personal
- Name Part
- Hayek, Karma
- Role
- Role Term:
Text
- creator
- Name:
Personal
- Name Part
- Uzun, Ece
- Role
- Role Term:
Text
- Advisor
- Name:
Personal
- Name Part
- Warner, Jeremy
- Role
- Role Term:
Text
- Reader
- Name:
Personal
- Name Part
- Chen, Elizabeth
- Role
- Role Term:
Text
- Reader
- Name:
Corporate
- Name Part
- Brown University. Biology and Medicine: Biotechnology
- Role
- Role Term:
Text
- sponsor
- Origin Information
- Copyright Date
- 2025
- Physical Description
- Extent
- , None p.
- digitalOrigin
- born digital
- Note:
thesis
- Thesis (Sc. M.)--Brown University, 2025
- Genre (aat)
- theses
- Abstract
- Despite advances in early detection and therapy, lung cancer is the leading cause of cancer death in the United States and accounts for approximately one-quarter of all cancer deaths. In early-stage non-small cell lung cancer (NSCLC), recurrence occurs in 30% to 55% of the patients, even after curative resection. In the event of recurrence, the 5-year survival after second operation decreases to 15%, indicating the poor prognosis with disease recurrence.. Lung adenocarcinoma (LUAD) patients account for a significant portion of recurred NSCLC cases, as LUAD is the most prevalent NSCLC subtype. In order to be able to offer more individualized treatment strategies for patients with LUAD, we developed two deep learning (DL) models tailored to predict binary (recurrence or no recurrence) and time-based (early vs. late recurrence i.e. before or after 12 months) LUAD recurrence by using clinical, mRNA, mutation, and methylation data.
To manage the high dimensionality of the multi-omics data, a modular feature selection pipeline was used, which integrated univariate filtering (ANOVA) and model-based methods (Logistic Regression, XGBoost, Random Forest) with hyperparameter optimization. The DL models were then built using Keras and benchmarked against conventional machine learning (ML) algorithms (K Neighbors Classifier, Logistic Regression, Naïve Bayes, Support Vector Machine, Random Forest, Decision Tree).
The binary model had AUROC 0.898 and AUPRC 0.867 with perfect precision but with compromised sensitivity, reflecting high predictive specificity but possible under-recognition of recurrence. The time-based model had an AUROC of 0.916 and an AUPRC of 0.956, with the detection of all true positives but over-estimation of false positives, indicating that further analysis is needed to obtain optimal model performance and generalizability. Further methods can be investigated in subsequent research to represent the complex, nonlinear relationships between omics layers (ex. graph neural networks, further hyperparameter tuning, etc.). Our method ultimately aims to investigate the usability of DL models to predict LUAD recurrence using clinical and genomic data from patients.
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01045739")
- Topic
- Oncology
- Subject
- Topic
- Deep Learning
- Language
- Language Term (ISO639-2B)
- English
- Record Information
- Record Content Source (marcorg)
- RPB
- Record Creation Date
(encoding="iso8601")
- 20250707