Related Experiment Videos
Identifying diagnostic errors with induced decision trees
1Penn State York, Pennsylvania 17403, USA. cxm53@psu.edu
Summary
Induced decision trees show higher accuracy in breast cancer diagnosis than pruned neural networks. Combining models can improve malignancy identification and describe ambiguous cases, enhancing diagnostic interpretation.
Area of Science:
- Medical informatics
- Machine learning in healthcare
- Oncology
Background:
- Accurate breast cancer diagnosis is crucial for effective treatment.
- Fine-needle aspirate readings are a key diagnostic tool.
- Interpreting diagnostic data can be challenging, leading to potential misclassifications.
Purpose of the Study:
- To compare the diagnostic accuracy of induced decision trees against pruned neural networks for breast cancer.
- To enhance the accuracy and interpretability of breast cancer diagnosis using fine-needle aspirate readings.
- To identify cases prone to misclassification by induced decision rules.
Main Methods:
- Utilized an online database of 699 suspected breast cancer cases.
- Induced decision trees from a randomly selected half of the data; tested accuracy on the remaining cases.
- Examined error patterns and created a dataset of misclassified cases to induce descriptive decision trees.
Main Results:
- Larger, less pruned decision trees demonstrated superior accuracy for both training and test data.
- Induced decision trees outperformed smaller pruned neural networks in diagnostic accuracy.
- Combining classifications from multiple trees improved identification of missed malignancies, though inductive error analysis did not reduce the overall error rate.
Conclusions:
- Overly compact models may miss minority class cases, such as malignancies.
- Combining diagnostic models and analyzing classification errors can improve breast cancer diagnosis.
- New methods can enhance malignancy detection and provide descriptions for ambiguous cases.