Related Experiment Videos
Diagnosis of breast cancer using Bayesian networks: a case study
Nicandro Cruz-Ramírez1, Héctor Gabriel Acosta-Mesa, Humberto Carrillo-Calvet
1Facultad de Física e Inteligencia Artificial, Universidad Veracruzana, Sebastián Camacho 5, Col. Centro, C. P. 91000 Xalapa, Veracruz, Mexico. ncruz@uv.mx
Computers in Biology and Medicine
|April 17, 2007
Summary
Bayesian network classifiers show promise for breast cancer diagnosis. However, observer subjectivity in analyzing fine-needle aspiration samples significantly reduces diagnostic accuracy, highlighting the need for standardized interpretation methods.
Area of Science:
- Medical Informatics
- Computational Biology
- Oncology
Background:
- Accurate breast cancer diagnosis is crucial for effective treatment.
- Bayesian network classifiers offer a computational approach to diagnostic challenges.
- Inter-observer variability in pathology can impact diagnostic tool performance.
Purpose of the Study:
- To assess the efficacy of seven Bayesian network classifiers for breast cancer diagnosis.
- To investigate the influence of observer subjectivity on classifier performance using real-world datasets.
Main Methods:
- Utilized two distinct real-world breast lesion fine-needle aspiration (FNA) databases.
- Employed seven different Bayesian network classifiers.
- Compared classifier performance on datasets collected by a single observer versus multiple observers.
Main Results:
- Achieved an average accuracy of 93.04% with single-observer data.
- Observed a reduced average accuracy of 83.31% with multi-observer data.
- Demonstrated a significant performance decrease due to inter-observer variability.
Conclusions:
- Bayesian network classifiers are potentially effective tools for breast cancer diagnosis.
- Subjectivity inherent in microscopic sample interpretation by different observers diminishes classifier accuracy.
- Standardization in pathological assessment is essential to maximize the utility of computational diagnostic tools in oncology.
