The Challenges of Applying Deep Learning for Hemangioma Lesion Segmentation

Alves, Pedro G.; Cardoso, Jaime S.; Bom-Sucesso, Maria

doi:10.1109/euvip.2018.8611730

Cited by 6 publications

(1 citation statement)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The authors of [12] tested five classic segmentation methods (Otsu, Canny, FCM, GVF snakes, and Shortest path) and two different convolutional neural network architectures, a standard linear CNN and a two-inputs DAG topology. The best results were obtained with the shortest path method and the two-inputs CNN.…”

Section: Introductionmentioning

confidence: 99%

Automatic Segmentation and Measurement of Infantile Hemangioma

2021

View full text Add to dashboard Cite

Infantile hemangiomas (IHs) are a type of vascular tumors that affect around 10% of newborns. The measurement of the lesion size and the assessment of the evolution is done manually by the physician. This paper presents an algorithm for the automatic computation of the IH lesion surface. The image scale is computed by using the Hough transform and the total variation. As pre-processing, a geometric correction step is included, which ensures that the lesions are viewed as perpendicular to the camera. The image segmentation is based on K-means clustering applied on a five-plane image; the five planes being selected from seven planes with the use of the Karhunen-Loeve transform. Two of the seven planes are 2D total variation filters, based on symmetrical kernels, designed to highlight the IH specific texture. The segmentation performance was assessed on 30 images, and a mean border error of 9.31% was obtained.

show abstract

Section: Introductionmentioning

confidence: 99%