Probing perfection: The relentless art of meddling for pulmonary airway segmentation from HRCT via a human-AI collaboration based active learning method

Shiyi Wang, Yang Nan, Sheng Zhang, Federico Felder, Xiaodan Xing, Yingying Fang, Javier Del Ser, Simon L. F. Walsh, Guang Yang

Abstract: Highlights•Application of Human-AI collaboration within our deep learning framework. Drastic reduction in the total amount of data to be labelled, leading to significant cost savings on annotation. Expert intervention in the machine learning training process allows for the efficient labelling of the most uncertain samples.•Utilization of the domain knowledge and experience of radiology experts for trachea annotation and correction of machine-generated tracheal centrelines. This addresses the issue of inaccurate intermediate results during initial iterations of model training.•Integration of domain knowledge from experts into our Human-Computer Interaction(HCI) based learning framework, resulting in a substantial enhancement of model performance.•Remarkable results with only 15%-35% of the training data when employing the Human-Computer Interaction based learning model, achieving performance equivalent to a supervised learning model (e.g., U-Net) that uses all available training data.•High flexibility and adaptability in the Human-Computer Interaction based learning framework, as network models and HCI query strategies can be replaced with various alternatives, offering robust reproducibility and diverse selection choices.