An optimized ensemble classifier for mammographic mass classification

Romesh Laishram

doi:10.1016/j.compeleceng.2024.109488

2024-10-01

PRE

Romesh Laishram *

Rinku Rabidas

DOI：10.1016/j.compeleceng.2024.109488

原文链接

原文求助

摘要

En 中文

Breast cancer is one of the most common cause of deaths among women due to cancer. A computer-assisted prognosis and diagnosis of breast cancer, which effectively reduces the mortality rate, is rapidly advancing field through medical imaging. Therefore, in an attempt, this paper introduces a novel deep learning and machine learning based hybrid framework for the classification of mammographic masses. Deep learning models are used to extract the deep features from the suspicious mass candidates obtained from a mass detection system while an optimized ensemble classifier is employed for the categorization of the detected mass lesions. The optimization is done via a nature-inspired tunicate swarm optimization where the cost function incorporates sensitivity and false positives to improved efficiency. Several experiments are performed with the proposed system and evaluated on 68 and 500 mammograms datasets from two standard databases-mini-MIAS and DDSM, respectively. In the experiment, false positive reduction, the model achieved sensitivities of 93.8% and 92.3% with false positives per image of 0.18 and 0.32, respectively for the mini-MIAS and DDSM datasets. For the characterization of malignant tumor against benign lesions and normal tissues, the introduced approach attained an A(z) value of 0.99 with an accuracy of 97.76% for the mini-MIAS dataset while the same for the DDSM dataset are 0.95 and 95.74%, respectively. The accuracy of a three-class categorization (normal, benign, and malignant) is 93.61% in the mini-MIAS dataset and 88.15% in the DDSM dataset, with malignancy F-1 scores of 0.86 and 0.80, respectively.

Keyword:

Breast Cancer diagnosis

Deep features

TSA

Ensemble classifier