Machine Learning and Deep Learning in Oncologic Imaging: Potential Hurdles, Opportunities for Improvement, and
Sireesha Yedururi1, Ajaykumar C Morani1, Venkata Subbiah Katabathina2
1From the Department of Abdominal Imaging, The University of Texas MD Anderson Cancer Center, Houston.
Journal of Computer Assisted Tomography
|July 16, 2021
Summary
Machine learning in oncologic imaging faces challenges like limited annotated data and inconsistent reporting. Solutions involve leveraging existing radiology reports to improve machine learning model development for cancer imaging analysis.
Area of Science:
- Radiology
- Oncology
- Machine Learning
- Artificial Intelligence
Background:
- Machine learning (ML) applications in clinical radiology, especially oncologic imaging, are rapidly advancing.
- Widespread implementation faces significant hurdles, including data scarcity and inconsistent reporting standards for solid tumors.
Purpose of the Study:
- To review potential challenges in applying ML to oncologic imaging.
- To identify opportunities for improvement and propose solutions for robust ML development.
Main Methods:
- Literature review focusing on ML in oncologic imaging.
- Analysis of common obstacles in data annotation and reporting consistency.
- Exploration of strategies to utilize existing radiology data.
Main Results:
- Key hurdles identified: insufficient annotated datasets and lack of standardized terminology/methodology.
- Opportunities exist in harmonizing reporting and data annotation practices.
- Radiology reports and annotations offer a valuable resource for ML model training.
Conclusions:
- Addressing data and standardization issues is crucial for ML adoption in oncologic imaging.
- Leveraging the wealth of data within radiology reports can accelerate ML development.
- Standardized approaches will enhance the reliability and applicability of ML tools in cancer imaging.
