The Evolving Synergy between AI and Human Expertise in Medical Diagnostics
Truong, Tri Dung (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025120432699
https://urn.fi/URN:NBN:fi:amk-2025120432699
Tiivistelmä
This thesis aims to systematically review and synthesize recent advances in artificial intelligence (AI) applications within medical imaging diagnostics, with a particular focus on collaborative frameworks in radiology that incorporate both human and AI capabilities. The objective is to identify foundational principles for designing interpretable and trustworthy AI-assisted diagnostic workflows.
A rigorous literature and systems review was conducted to examine current research on collaborative diagnostic systems in medical imaging. Special attention was paid to human-in-the-Loop (HITL) frameworks and explainable AI (XAI) approaches that enhance clinician understanding and decision-making. The review evaluated how AI assistance affects diagnostic accuracy, efficiency, and clinician confidence.
The study found that deep learning-based AI systems have achieved near-expert performance in disease classification, lesion detection, and organ segmentation. However, challenges related to interpretability, trust, and generalizability hinder widespread clinical adoption. While AI can significantly augment clinical practice, persistent gaps remain, including limited real-world HITL evaluations and underutilization of uncertainty quantification techniques. The synthesis established evidence-based guidelines for the development of safe, robust, and interpretable AI-assisted diagnostic workflows.
The review highlights a need for more real-world evaluations of collaborative systems and greater emphasis on uncertainty quantification in future research. The limited availability of comprehensive HITL frameworks in the current literature restricts the generalizability of findings.
The insights provided serve as a guide for the practical implementation of AI in clinical radiology. Emphasizing collaboration between clinicians and AI, the study informs the design of diagnostic systems that enhance accuracy and clinician confidence while ensuring safety and interpretability.
This thesis offers a structured synthesis of the literature on AI–human collaboration in medical imaging diagnostics, addressing existing challenges and proposing foundational principles for future research and practice. It positions AI as a supportive partner in clinical workflows, thereby advancing the safe and effective integration of AI technologies in healthcare settings.
A rigorous literature and systems review was conducted to examine current research on collaborative diagnostic systems in medical imaging. Special attention was paid to human-in-the-Loop (HITL) frameworks and explainable AI (XAI) approaches that enhance clinician understanding and decision-making. The review evaluated how AI assistance affects diagnostic accuracy, efficiency, and clinician confidence.
The study found that deep learning-based AI systems have achieved near-expert performance in disease classification, lesion detection, and organ segmentation. However, challenges related to interpretability, trust, and generalizability hinder widespread clinical adoption. While AI can significantly augment clinical practice, persistent gaps remain, including limited real-world HITL evaluations and underutilization of uncertainty quantification techniques. The synthesis established evidence-based guidelines for the development of safe, robust, and interpretable AI-assisted diagnostic workflows.
The review highlights a need for more real-world evaluations of collaborative systems and greater emphasis on uncertainty quantification in future research. The limited availability of comprehensive HITL frameworks in the current literature restricts the generalizability of findings.
The insights provided serve as a guide for the practical implementation of AI in clinical radiology. Emphasizing collaboration between clinicians and AI, the study informs the design of diagnostic systems that enhance accuracy and clinician confidence while ensuring safety and interpretability.
This thesis offers a structured synthesis of the literature on AI–human collaboration in medical imaging diagnostics, addressing existing challenges and proposing foundational principles for future research and practice. It positions AI as a supportive partner in clinical workflows, thereby advancing the safe and effective integration of AI technologies in healthcare settings.