A survey on active learning and human-in-the-loop deep learning for medical image analysis

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning Applications

Background:

• Deep learning (DL) is state-of-the-art for medical image analysis, aiding in detection, diagnosis, and treatment planning.
• However, the complexity of medical data necessitates retaining human oversight in DL systems for safety and efficacy.
• Human-in-the-Loop (HITL) computing is vital for integrating human expertise into DL workflows within the safety-critical medical domain.

Purpose of the Study:

• To review the role of humans in developing and deploying DL-enabled diagnostic applications.
• To focus on techniques that ensure significant human end-user input in clinical DL systems.
• To explore the future of HITL computing in medical AI.

Main Methods:

• Review of key areas vital for DL in clinical practice, emphasizing human-AI interaction.
• Evaluation of four critical components: Active Learning for data annotation, interaction with model outputs via iterative feedback, practical deployment considerations, and future research directions.
• Analysis of knowledge gaps and related fields that can advance HITL computing.

Main Results:

• Active Learning strategies can optimize model performance by selecting the most informative data for annotation.
• Iterative feedback mechanisms allow users to steer model predictions and interpret results meaningfully.
• Practical considerations and future research are essential for the successful clinical deployment of HITL DL systems.

Conclusions:

• HITL computing is crucial for advancing the safe and effective use of deep learning in clinical diagnostics.
• Integrating human expertise through active learning and interactive feedback enhances model performance and interpretability.
• Further research unifying various HITL aspects is needed to realize the full potential of AI in medical imaging.