ICHNO-ECHNO 2022 - Abstract Book

S34

ICHNO-ECHNO 2022

The original Unet configuration as presented in was used as the DL architecture.

Figure 1 shows an example of CT image before and after processing with the SLF for the neck. In the right image, both bones and soft muscular tissue have a better contrast.

Results The Unet with the SLF preprocessing and hard region weighted loss performed the best, reaching around 77% validation Dice scores as seen from figure 2. The Unet trained with Dice performed the second best, reaching around 70% validation Dice scores. The Unet trained with generalized surface distance loss performed the worst, reaching around 58% for the best validation model. However, this model was able to classify smaller organs like the lens better than the other models, reaching about 100% in some samples.

Conclusion Despite leading to the poorest overall results in terms of predicted Dice scores, our proposed loss function allowed the model to segment small organs like the lens more accurately. In future a combination of Dice score and the surface.