Abdominal organ segmentation on ultrasound

A medtech company developing an assisted-guidance ultrasound solution needed pixel-precise segmentations of the main abdominal organs on 2D ultrasound frames and video clips, with the specific challenges of speckle, low contrast and operator-dependent views.

Produce binary and multi-class masks for the liver, kidneys, pancreas, abdominal aorta, gallbladder and spleen, delivered in a format compatible with the client's training pipeline.

Radiologists experienced with abdominal ultrasound were trained on the client's protocol and used an open-source, browser-based annotation platform. Guidelines emphasized annotation consistency, use of anatomical landmarks and skipping frames where visibility was insufficient.

The client received a homogeneous dataset across a broad range of operators and acquisition conditions. Regular sync sessions with the annotators helped refine borderline cases directly and kept quality high across a demanding modality.