Combination of enhanced AdaBoosting techniques for the characterization of breast cancer tumors

Breast cancer is a life threatening disease affecting one of every eight women, with the risk increasing significantly with age. Non-invasive diagnostic modalities are preferable over biopsy. However, these non-invasive diagnostic techniques have lower diagnostic accuracy when compared to biopsy. Currently, CAD (Computer-Aided Diagnosis) techniques have demonstrated strong potential to increase the accuracy of such non-invasive diagnostic modalities. This paper proposes the use of boosting to increase the accuracy of CAD-based techniques to solve the breast cancer characterization problem. It investigates and compares the performance of popular variants of the boosting algorithms, namely: AdaBoosting, AveBoosting, GentleBoosting, ConserBoosting, and Average Conservative Boosting and their suitability for the breast cancer tumor characterization problem. This work also proposes a hybrid boosting algorithm that combines the advantages of several boosting techniques. The results of applying the different boosting techniques investigated on real breast cancer benchmarks show that the hybrid boosting algorithm outperforms the other boosting techniques on average by 48%.