Multitask Deep Neural Network for the Fully Automatic Measurement of the Angle of Progression

Figure 5

Summary of different methods. These methods include Unet, MT-Unet_A, MT-Unet_B, MT-Unet_C, MT-Unet_D, and MT-Unet. (a) Comparison of the prediction results of different segmentation methods. (b) Comparison of the prediction results of different location methods. (c) Comparison of AoP results of different methods. Compared with MT-Unet, MT-Unet_D is without SLF, MT-Unet_C is without SLF and AFM, MT-Unet_B is without SLF and ECA, MT-Unet_A is without SLF, AFM, and ECA. Note: Dist_L/Dist_R denotes the Euclidean distances between predicted left/right endpoints and true left/right endpoints. APT denotes the angle between the true line and the predicted line of two endpoints. AoP denotes the absolute value of the AoP difference between the true AoP and the predicted one (complete and detailed results are provided in Appendix S3).

(a) Image segmentation (Task1)

(b) Endpoint detection (Task2)

(c) AoP calculations