Table 7. Classification performance of different methods.
| Characteristics | Malignancy (%) | Margin (%) | Texture (%) | Calcification (%) |
|---|---|---|---|---|
| Methods | Acc/Sens/Spec | Acc/Sens/Spec | Acc/Sens/Spec | Acc/Sens/Spec |
| Encoding + RF | 50.3/51.6/46.6 | 52.7/49.2/57.2 | 58.5/71.5/31.0 | 68.7/88.6/22.4 |
| Decoding + RF | 75.2/66.3/79.0 | 53.2/52.9/53.6 | 63.4/70.5/49.0 | 65.4/51.1/70.0 |
| Decoding + residuals | 73.1/70.6/78.2 | 74.6/78.2/69.9 | 83.2/92.0/70.0 | 84.0/87.2/65.0 |
| Encoding + residuals | 75.6/71.8/80.0 | 73.4/79.7/65.4 | 80.0/89.4/64.8 | 86.9/93.4/69.0 |
| Proposed | 77.8/83.4/76.2 | 75.6/81.4/75.0 | 85.2/92.7/78.2 | 87.1/92.4/80.0 |
Encoding + RF/decoding + RF represent the encoding/decoding feature coupled with a random forest; Decoding + residuals/encoding + residuals represent the single branch (encoding/decoding) network; Acc, Sens, and Spec represent the accuracy, sensitivity, and specificity of each network.