Multi-task weak supervision enables anatomically-resolved abnormality detection in whole-body FDG-PET/CT.

Area of Science:

• Medical Imaging Analysis
• Machine Learning in Healthcare
• Natural Language Processing for Clinical Data

Background:

• Whole-body FDG-PET/CT imaging offers clinical value but faces challenges in applying supervised machine learning due to limited labeled data and large exam sizes.
• Existing supervised learning methods struggle with the complexity and data requirements of whole-body PET/CT scans.

Purpose of the Study:

• To develop a weak supervision framework for extracting granular regional abnormality labels from free-text radiology reports.
• To train an attention-based, multi-task Convolutional Neural Network (CNN) architecture for abnormality detection and location estimation in whole-body scans.
• To assess the framework's utility in improving performance on rare abnormalities and enhancing mortality prediction.

Main Methods:

• Leveraging natural language processing to extract imperfect, granular regional abnormality labels from unstructured radiology reports.
• Developing a custom ontology for labeling anatomical regions and creating structured pathology profiles for each imaging exam.
• Training an attention-based, multi-task CNN using the weakly supervised labels for abnormality detection and location estimation.

Main Results:

• The weak supervision framework successfully generated structured pathology profiles from free-text reports.
• The multi-task CNN trained on these labels demonstrated strong performance in detecting and locating abnormalities, especially rare ones.
• The learned representation improved the accuracy of mortality prediction from imaging data.

Conclusions:

• Weak supervision using NLP on radiology reports is a viable strategy to overcome data limitations in whole-body FDG-PET/CT analysis.
• The proposed framework enhances machine learning model performance for abnormality detection and has potential applications in prognostic prediction.
• This approach offers a scalable solution for leveraging clinical text data to improve medical imaging interpretation.