Related Experiment Video
Updated: Nov 15, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
AI applications to medical images: From machine learning to deep learning
Isabella Castiglioni1, Leonardo Rundo2, Marina Codari3
1Department of Physics, Università degli Studi di Milano-Bicocca, Piazza della Scienza 3, 20126 Milano, Italy; Institute of Biomedical Imaging and Physiology, National Research Council, Via Fratelli Cervi 93, 20090 Segrate, Italy.
Summary
This review addresses challenges in developing artificial intelligence (AI) clinical decision support systems. Clarifying these issues is crucial for advancing AI in biomedical research and healthcare services.
Area of Science:
- Biomedical research and healthcare services
- Artificial intelligence applications in medicine
- Medical imaging analysis
Background:
- Artificial intelligence (AI) models are increasingly integrated into biomedical research and healthcare.
- Clinical decision support systems (CDSS) are a key area for AI implementation.
- Real-world application of AI in healthcare presents unique development challenges.
Purpose of the Study:
- To review and clarify critical challenges in developing AI applications for clinical decision support systems.
- To provide insights into the practical implementation of AI in real-world healthcare contexts.
- To guide researchers and developers in overcoming obstacles in AI translation to clinical practice.
Main Methods:
- A narrative review methodology was employed.
- A critical assessment of articles published between 1989 and 2021 was conducted.
- Key challenges in AI development for clinical decision support were identified and analyzed.
Main Results:
- Illustrates architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches.
- Details data curation steps including labelling, annotation, harmonization, and federated learning.
- Discusses sample size calculation, data augmentation, and AI model interpretability (black box issue).
Conclusions:
- Biomedicine and medical imaging are highly promising domains for AI applications.
- Addressing specific challenges is essential for successful AI system development and clinical translation.
- Clarification of development hurdles facilitates the integration of AI into healthcare practices.
