Oncology research · 2026
Exploring Machine Learning Approaches for Decision Support in Neoadjuvant Therapy of Locally Advanced Rectal Cancer.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Dhar Eshita, Kabir Muhammad Ashad, Nadar Divyabharathy Ramesh, Kuo Li-Jen, Jonnagaddala Jitendra, Huang Yaoru, Uddin Mohy, Syed-Abdul Shabbir
Аннотация
OBJECTIVES: Decisions regarding CT after nCCRT for locally advanced rectal cancer (LARC) are challenging due to limited evidence guiding treatment. This study aimed to (i) evaluate the predictive performance of machine learning (ML) models in patients treated with neoadjuvant concurrent chemoradiotherapy (nCCRT) alone vs. those receiving nCCRT plus chemotherapy (CT), (ii) identify features associated with treatment improvement, and (iii) derive ML-based thresholds for treatment response.
METHODS: This retrospective study included 409 patients with LARC treated at three affiliated hospitals of Taipei Medical University. Patients were categorised into two groups: nCCRT alone followed by surgery (n = 182) and nCCRT plus additional CT (n = 227). Thirty-four baseline demographic, tumor, and laboratory variables were analysed. Four ML algorithms (K-Star, Random Forest, Multilayer Perceptron, and Random Committee) were evaluated, while five feature-ranking algorithms identified influential attributes among improved patients across both treatments. Decision Stump and AdaBoostM1 were applied to derive threshold-based patterns.
RESULTS: K-Star achieved the highest accuracy for nCCRT alone (80.8%; AUC = 0.89), while Random Committee performed best for nCCRT plus CT (77.3%; AUC = 0.84). Clinical N stage (cN) ranked highest, followed by Sodium (Na), Glutamic pyruvic transaminase, estimated glomerular filtration rate, body weight, red blood cell count, mean corpuscular hemoglobin concentration, and blood urea nitrogen. Threshold patterns suggested that CT-related improvement aligned with higher lymphocyte percentage and lower platelet distribution width, whereas nCCRT-only improvement aligned with elevated eGFR, GPT, and cN = 2. Conclusions: ML-based analysis identified key predictors and demonstrated good model performance, supporting individualised post-nCCRT chemotherapy decisions.
METHODS: This retrospective study included 409 patients with LARC treated at three affiliated hospitals of Taipei Medical University. Patients were categorised into two groups: nCCRT alone followed by surgery (n = 182) and nCCRT plus additional CT (n = 227). Thirty-four baseline demographic, tumor, and laboratory variables were analysed. Four ML algorithms (K-Star, Random Forest, Multilayer Perceptron, and Random Committee) were evaluated, while five feature-ranking algorithms identified influential attributes among improved patients across both treatments. Decision Stump and AdaBoostM1 were applied to derive threshold-based patterns.
RESULTS: K-Star achieved the highest accuracy for nCCRT alone (80.8%; AUC = 0.89), while Random Committee performed best for nCCRT plus CT (77.3%; AUC = 0.84). Clinical N stage (cN) ranked highest, followed by Sodium (Na), Glutamic pyruvic transaminase, estimated glomerular filtration rate, body weight, red blood cell count, mean corpuscular hemoglobin concentration, and blood urea nitrogen. Threshold patterns suggested that CT-related improvement aligned with higher lymphocyte percentage and lower platelet distribution width, whereas nCCRT-only improvement aligned with elevated eGFR, GPT, and cN = 2. Conclusions: ML-based analysis identified key predictors and demonstrated good model performance, supporting individualised post-nCCRT chemotherapy decisions.