Fig 1.
Illustrates the flowchart of the study process.
Fig 2.
Overview of feature based transfer learning framework.
Table 1.
Demographic data for 294 patients.
Table 2.
Classification performance of transfer learning models and radiologists.
Fig 3.
Confusion matrices of transfer learning models and radiologists on the test set.
Table 3.
Performance comparison of transfer learning models versus radiologists in breast mass classification.
Fig 4.
Receiver operating characteristic curves of transfer learning models and radiologists on the test set.
Fig 5.
Training and test set curves for each model.
Fig 6.
Figures A and C show B-mode ultrasound images of a fibroadenoma and an inflammatory masses, respectively, from the test set, and Figures C and D show the corresponding images after overlaying the thermograms using the pre-trained DenseNet121 model.
Fig 7.
Figures A and C show the B-mode ultrasound images of invasive ductal carcinoma and mucinous carcinoma in the test set, respectively, and Figures C and D show the respective images after overlaying the thermograms using the pre-trained DenseNet121 model.