18 June 2020 : Clinical Research
Ensemble Deep Learning Model for Multicenter Classification of Thyroid Nodules on Ultrasound Images
Xi Wei1ACDEG*, Ming Gao2BEF, Ruiguo Yu3BCD, Zhiqiang Liu3BCE, Qing Gu4BC, Xun Liu5BC, Zhiming Zheng6BC, Xiangqian Zheng2BC, Jialin Zhu1ABCDEF, Sheng Zhang1BDDOI: 10.12659/MSM.926096
Med Sci Monit 2020; 26:e926096
Supplementary Table 3 Classification algorithm structure.
| Layer | Detail | Output size |
|---|---|---|
| Convolution | 3×3 conv | 64×64×16 |
| Dense Block1 | {3×3 conv }×17 | 64×64×220 |
| Transition Layer1 | 1×1 conv | 32×32×220 |
| 2×2 avg pool | ||
| Dense Block2 | {3×3 conv }×17 | 32×32×424 |
| Transition Layer2 | 1×1 conv | 16×16×424 |
| 2×2 avg pool | ||
| Dense Block3 | {3×3 conv }×17 | 16×16×628 |
| Transition Layer3 | 1×1 conv | 8×8×628 |
| 2×2 avg pool | ||
| Dense Block4 | {3×3 conv }×17 | 8×8×832 |
| Transition Layer4 | 1×1 conv | 4×4×832 |
| 2×2 avg pool | ||
| Dense Block5 | {3×3 conv }×17 | 4×4×1036 |
| Batch Normalization | 4×4×1036 | |
| Relu | 4×4×1036 | |
| Pooling | 4×4 avg pool | 1×1×1036 |
| Fully Connection | 1036 | |
| Fully Connection | 2 | |
| Softmax | 2 |