BACKGROUND
Artificial intelligence can help physicians improve the accuracy of breast cancer diagnosis. However, the effectiveness of AI applications is limited by doctors’ adoption of the results recommended by the AI systems. A case-based reasoning system for breast cancer diagnosis (CBR-BCD) that considers the effects of external characteristics of cases (ECC) can not only provide doctors with more accurate results for auxiliary diagnosis, but also improve doctors’ trust in the results, so as to encourage doctors to adopt the results recommended by the system.
OBJECTIVE
The objective of our study is to develop a novel integrated case-based reasoning (CBR) framework based on Naive Bayes and K-Nearest Neighbor (KNN) algorithms considering the effects of external characteristics of cases (CBR-ECC) and a corresponding system named CBR-BCD to assist in diagnosis and promote adoption by doctors.
METHODS
We used a real-world data set from the Maputo Central Hospital in Mozambique and constructed the CBR-ECC model and corresponding CBR-BCD system. We performed data processing and obtained six internal features and three external features of the cases. We randomly divided the 1214 cases into a training group and a testing group. The performance of the model was evaluated by accuracy and the area under the receiver operating characteristic curve (AUC).
RESULTS
The system based on the CBR-ECC model was developed. In the first stage of this model, Naive Bayes showed the best performance, compared with KNN and J48 decision tree classifiers, with an accuracy rate of 95.87%. In the second stage, the accuracy of the KNN model with the optimal K value of 2 was 99.40%. In the third stage, after considering the external characteristics of the cases, the rankings of recommendation changed. Finally, we report the users’ evaluation of the novel CBR system in a real hospital scenario; we found that it is superior to the original system.
CONCLUSIONS
CBR-BCD not only enables accurate case recommendations to support health practitioners in diagnosing breast cancer and reducing diagnostic inaccuracies, but also facilitates the adoption of system-recommended results by physicians, which is valuable for clinicians to assist in diagnosis. It enables the early screening of breast cancer to improve the quality of breast cancer management and reduces the socioeconomic burden compared to traditional methods.