Amplifying Human Capabilities in Prostate Cancer Diagnosis: An Empirical Study of Current Practices and AI Potentials in Radiology

Title of the Paper

Enhancing Human Capability in Prostate Cancer Diagnosis: An Empirical Study of Current Practices and the Potential of Artificial Intelligence in Radiology

Paper Information

• Subject Area: Radiology and Artificial Intelligence (AI), with a focus on prostate cancer diagnosis
• Keywords: Prostate cancer diagnosis, Human-Centered Artificial Intelligence (HCAI), Contextual Inquiry, Radiologists, Computer-Aided Detection (CAD), Deep Learning (DL), XiangAI, Multidisciplinary Collaboration

Research Background and Issues

• Identified Problems or Challenges:

  1. Prostate cancer is the second most common cancer among men worldwide, with cases expected to increase annually, particularly in Germany.
  2. There is a shortage of radiologists, creating a conflict between diagnostic demand and clinical workload.
  3. While AI holds great potential, its practical application in medical practice remains limited, especially due to the lack of integration between human practices and technology.
  4. Repetitive and manual tasks are still prevalent, making diagnostic workflows prone to human error, particularly in multiparametric MRI diagnosis based on PI-RADS (Prostate Imaging Reporting and Data System) scoring, such as volume measurement and PSA density calculation.

• Significance: Early detection and accurate diagnosis of prostate cancer are critical for reducing disease severity. Improving diagnostic efficiency and reducing radiologists' workload can enhance patient outcomes.

• Motivation and Related Work:

  • Previous studies have shown that integrating AI into radiology can improve accuracy, efficiency, and consistency.
  • Many studies have explored AI applications in diagnosing breast cancer and other diseases, but there is a lack of focus on human-AI collaboration and practical integration.
  • Adopting an HCAI approach ensures that technology design aligns more closely with real-world practice needs.

Proposed Solution

• Proposed Solution:

  • Combining Human-Centered Artificial Intelligence (HCAI) with practice-centered design to enhance prostate cancer diagnosis through human-AI collaboration and technological assistance.
  • Proposing an AI implementation framework named PAIRADS to support MRI image analysis and diagnostic tasks, such as prostate region segmentation and lesion detection.

• Innovations:

  1. Incorporating radiologists' practical expertise into AI design and addressing existing challenges through collaborative interaction.
  2. Introducing a dual-diagnosis model: positioning AI as an assistive tool for radiologists rather than a replacement for human judgment.
  3. Emphasizing interpersonal and multidisciplinary collaboration: connecting radiologists, urologists, and pathologists via an automated feedback system to enable diagnostic data sharing and validation.

• Implementation Steps and Key Technologies:

  1. Field Research: Conducting contextual inquiries and in-depth interviews at four radiology centers in Germany to gather insights into diagnostic practices and challenges.
  2. AI System Design: Training AI models using semi-supervised learning (SSL) and deep convolutional neural networks (DCNN) to learn region segmentation and lesion detection from MRI images.
  3. Prototype Development: Developing AI-driven visualization tools for automatic region segmentation, lesion highlighting, and PI-RADS classification.
  4. Automation of Tasks: Introducing automated PSA density calculation and report generation to reduce radiologists' manual workload.
  5. Facilitating Cross-Disciplinary Communication: Building a multidisciplinary platform to streamline information exchange and feedback mechanisms.

Research Outcomes

• Specific Findings:

  1. A structured analysis of current practices (including diagnostic workflows, methods, and systems) revealed five main diagnostic stages: patient inquiry, image acquisition, image interpretation, report generation, and validation.
  2. Proposed a design framework for prostate cancer diagnosis, including AI-based visualization support tools and automation features.
  3. Highlighted the importance of "hybrid intelligence" to improve diagnostic accuracy and proposed a collaborative framework between radiologists and AI.

• Advantages:

  1. Accelerated diagnostic speed: Automated volume measurement and PI-RADS scoring reduce human error.
  2. Improved diagnostic consistency: Minimizes the impact of experience variability on outcomes.
  3. Reduced workload: Automates common repetitive tasks (e.g., report generation and data transfer).

• Experimental or Evaluation Results:

  • Although the AI system in the study has not been widely implemented, initial prototypes, such as the prostate boundary segmentation model, have shown preliminary feasibility.
  • Radiologists expressed a generally positive attitude toward AI, viewing it as an assistive tool rather than a standalone diagnostic decision-maker.

• Limitations and Future Directions:

  1. Limited data sample size; larger-scale real patient data is needed to validate the robustness of the AI model.
  2. The current AI system is still in the experimental phase and requires further refinement to adapt to diverse medical environments.
  3. Future steps include real-world clinical application testing to evaluate the efficiency and trustworthiness of the integrated AI system.