UNIPS - ESTS – BIBLIOTECA
URI permanente desta comunidade:
Navegar
Percorrer UNIPS - ESTS – BIBLIOTECA por orientador "Albuquerque, Joana"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Machine learning-based assessment of mutational profiles in lung carcinomaPublication . Graúdo, João Rafael Vieira; López, Miguel; Albuquerque, JoanaThis study aimed to develop machine learning models for automated prediction of overall survival and mutational status in advanced non-small cell lung cancer (NSCLC) patients with actionable molecular alterations. A retrospective cohort of 275 stage IV NSCLC patients from five hospitals in Southern Portugal (2016–2021) was analysed. Clinical, demographic, molecular, and therapeutic data were integrated into four supervised classification algorithms: Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and AdaBoost. Models were optimized using 10-fold stratified cross-validation with hyperparameter tuning via Grid Search. To address class imbalance, three complementary approaches were implemented: baseline modeling, Synthetic Minority Over-sampling Technique (SMOTE), and synthetic data augmentation using Conditional Tabular Generative Adversarial Network (CTGANSynthesizer). Performance was evaluated using accuracy, precision, recall, F1-score, and Area Under the Curve – Receiver Operating Characteristics (AUC-ROC) metrics. Results demonstrated that ensemble-based methods (RF, XGBoost, and AdaBoost) substantially outperformed SVM, particularly when trained on balanced datasets. The third approach, incorporating both synthetic data generation and SMOTE oversampling, yielded the highest discriminatory performance, with AdaBoost achieving an AUC-ROC of 0.9217. Correlation analysis revealed that Eastern Cooperative Oncology Group Performance Status (r=0.220), bone metastases (r=0.179), and sex (r=0.159) were the strongest positive predictors of mortality, while Epidermal Growth Factor Receptor (EGFR) exon 19 deletions (r=-0.140) demonstrated the most favorable prognostic association. The most prevalent molecular alteration was Kirsten Rat Sarcoma Virus (KRAS) G12C (35.64%), followed by EGFR mutations (14.91%) and Anaplastic Lymphoma Kinase (ALK) rearrangements (7.27%), consistent with European epidemiological data. This work demonstrates how machine learning tools can be valuable in predicting survival outcomes and personalizing treatments for patients with advanced NSCLC. Despite these advantages, there are still important challenges, such as the issue of data imbalance and the need to validate models in independent patient groups. Therefore, it is essential to maintain rigorous methodologies throughout the process. In the future, it will be important to test these models in different hospitals, integrate imaging data, and develop decision-support tools that are simple and transparent for healthcare professionals to use in their daily practice.