Pairwise machine learning-based automatic diagnostic platform utilizing CT images and clinical information for
An-du Zhang1, Qing-Lei Shi2,3, Hong-Tao Zhang4
1Department of Radiotherapy, Hebei Medical University Fourth Affiliated Hospital and Hebei Provincial Tumor Hospital, 12 Jiankang Road, Shijiazhuang, Hebei, 050011, People's Republic of China.
Abdominal Radiology (New York)
|June 3, 2024
Summary
A machine learning model accurately predicts locoregional recurrence (LR) in elderly esophageal squamous cell cancer (ESCC) patients treated with radiotherapy. This tool shows promise for an automated esophageal cancer diagnosis system, improving clinical practice.
Area of Science:
- Oncology
- Radiotherapy
- Machine Learning
Background:
- Elderly patients with esophageal squamous cell cancer (ESCC) undergoing radical radiotherapy face risks of locoregional recurrence (LR).
- Accurate prediction of LR is crucial for optimizing treatment strategies and improving patient outcomes in this demographic.
Purpose of the Study:
- To assess the feasibility and accuracy of a machine learning algorithm in predicting LR in elderly ESCC patients post-radiotherapy.
- To develop a predictive model utilizing radiomics and clinical factors for early identification of recurrence risk.
Main Methods:
- A pairwise naive Bayes (NB) model was developed using 130 patient datasets, divided into training (70%) and testing (30%) sets.
- Radiomics features were extracted from pretreatment CT scans, combined with clinical factors, and analyzed using pyradiomics software.
- Model performance was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis (DCA).
Main Results:
- The study identified 10 radiomics features and 2 clinical factors significant for LR prediction.
- The NB model achieved high prediction performance, with an area under the ROC curve of 0.944 in the testing cohort and an accuracy of 0.914.
- The model demonstrated good calibration and clinical validity, with a positive predictive value of 85.71% for LR in the testing cohort.
Conclusions:
- A pairwise NB model integrating radiomics and clinical data can accurately predict LR in elderly ESCC patients after radical radiotherapy.
- The developed model shows significant potential for integration into an automated esophageal cancer diagnostic system for clinical application.
- This approach offers a promising tool for personalized risk assessment and management of locoregional recurrence in ESCC.
