TransNet: a comparative study on breast carcinoma diagnosis with classical machine learning and transfer learning paradigm

• Published in: Multimedia tools and applications 2024-03, Vol.83 (11), p.33855-33877
• Main Authors: Chugh, Gunjan, Kumar, Shailender, Singh, Nanhay
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Breast; Breast cancer; Classifiers; Comparative studies; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Deep learning; Diagnosis; Digital imaging; Feature extraction; Machine learning; Medical imaging; Multimedia Information Systems; Neural networks; Special Purpose and Application-Based Systems; Support vector machines; Track 2: Medical Applications of Multimedia
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-16938-x

Breast Carcinoma is a deadly disease; therefore, timely diagnosis is one of the most critical concerns that must be addressed globally since it can significantly enhance overall survival rates. Currently, Medical Imaging relies on Machine Learning(ML) and Deep Learning(DL) for accurate and early identification of diseases. In this article, a framework is proposed for diagnosing & classifying breast tumors using deep learning approaches. We have performed two experiments on the CBIS-DDSM (Curated Breast Imaging Subset of Digital Database for Screening Mammography) dataset. In the first approach, i.e., Deep feature extraction with ML classifier head, Deep Convolutional Neural Network(DCNN) models such as VGG16, VGG19, Res Net 50, and Res Net 152 are deployed as feature extractors, and the obtained features are utilized for training conventional machine learning classifiers. The second approach, called Deep Learning feature extraction with a neural network classifier, exploits Mobile Net, VGG16, VGG19, ResNet50, Res Net 152, and, Dense Net 169 for feature extraction and categorization. The results show that in the first case, Random Forest (RF) and XG Boost (XGB) Classifier perform best with 100% accuracy on the training set, whereas Support Vector Machine (SVM) and XGB exhibit 95%(+-5%) on the Test dataset for all the models. In the second approach, Mobile Net, ResNet50, and Dense Net 169 outperform the other models with an accuracy of 97%(+-2%) for both the Training and Test sets. The evaluated results have shown that the second approach depicts an increase in accuracy by 4%.