The transformational potential of molecular radiomics

Area of Science:

• Nuclear Medicine
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Conventional radiomics relies on predefined regions of interest.
• Artificial intelligence (AI) enhances feature extraction in radiomics.
• Deep learning (DL) enables direct feature extraction from images, bypassing segmentation.

Purpose of the Study:

• To elucidate the principles of radiomics and deep radiomics in nuclear medicine.
• To bridge the understanding gap regarding deep molecular radiomics.
• To provide a foundational understanding for AI-augmented nuclear medicine.

Main Methods:

• Exploration of traditional radiomic feature extraction methods.
• Introduction to AI-augmented segmentation and feature extraction.
• Explanation of deep learning (DL) for abstract radiomic feature extraction (deep radiomics).

Main Results:

• Deep radiomics extracts high-dimensional, abstract features directly from image tensors.
• Molecular radiomics, including deep molecular radiomics, offers insights beyond conventional imaging quantitation.
• Deep radiomics is well-suited for the high data density in hybrid imaging.

Conclusions:

• Deep radiomics represents a significant advancement over traditional radiomics in nuclear medicine.
• Understanding deep radiomics is crucial for its broader acceptance and integration into clinical decision-making.
• This work aims to demystify the concepts of radiomics and deep radiomics for nuclear medicine practitioners.