EfficientDeepLab for Automated Trachea Segmentation on Medical Images

• Arthur Guilherme Santos Fernandes UFMA
• Geraldo Braz Junior UFMA
• João Otávio Bandeira Diniz UFMA
• Aristófanes Correa Silva UFMA
• Caio Eduardo Falcõ Matos UFMA

Resumo

Segmentation of Organs at Risk is a fundamental step during radiotherapy planning for cancer treatment. Its goal is to preserve healthy tissue around the tumor and ensure that the most radiation strikes only cancer cells. Physicians do this job manually, which can be slow and error-prone. Thus, automatic segmentation methodologies can speed up organ delimiting during radiotherapy planning. This work designs a method, EfficientDeepLab, a convolutional neural network architecture trained on CT scans for trachea segmentation, and obtained an 88.6% dice score.