Active guidance in ultrasound bladder scanning using reinforcement learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Urology

Background:

• Accurate bladder volume measurement is crucial for diagnosing urinary retention and voiding dysfunction.
• Optimal ultrasound imaging can be challenging for less experienced operators, potentially leading to misdiagnosis.

Purpose of the Study:

• To develop an intelligent guidance system using reinforcement learning (RL) to improve ultrasound image acquisition in bladder scanning.
• To enhance image quality and consistency in bladder assessments.

Main Methods:

• Implementation of a novel pipeline with a variant of Deep Q-Networks (DQN) called Adam LMCDQN.
• Development of a domain-specific reward design for RL, incorporating domain knowledge.
• Validation within linear Markov Decision Processes in a simulated environment.

Main Results:

• Achieved a promising success rate in reaching target points: [Formula: see text] (transverse) and [Formula: see text] (longitudinal).
• Significantly outperformed supervised deep learning models, which achieved [Formula: see text] (transverse) and [Formula: see text] (longitudinal).
• Demonstrated technical feasibility of optimal-view localization using RL.

Conclusions:

• Reinforcement learning shows significant potential for improving ultrasound bladder assessment.
• The proposed intelligent guidance system offers real-time, adaptive feedback to operators.
• This work is a pioneering application of RL in ultrasound guidance for bladder assessment.