Abstract
P873
Introduction: Radiomics and AI-assisted medical image analysis are growing areas of great interest. Many development frameworks are available for conducting radiomic and AI-based image analysis, yet most of them require some programming knowledge. In 2017, the LIFEx software was introduced as a user-friendly platform to support medical image display, basic image processing and automated calculation of radiomic features without requiring any coding skills (Nioche et al, Cancer Res 2018). Here, we present some of the many new functionalities that have been included in LIFEx over the past five years to make it an up-to-date and practical tool for efficiently performing radiomic and AI studies.
Methods: LIFEx evolutions have been driven by 1) the need to comply with the Image Biomarker Standardisation Initiative (IBSI) guidelines (Zwanenburg et al, Radiology 2020), 2) a careful follow-up of advances in the field, 3) new needs in terms of image annotation for supervised learning, 4) fruitful interactions with LIFEx users. The correct implementation of radiomic feature calculation has been thoroughly checked using IBSI benchmarks. Novel experimental and validated radiomic features have been implemented. A practical annotation module has been developed.
Results: LIFEx is independent of any commercial libraries, is written in Java, and runs on Windows, MacOS, and Linux after a few-click installation process. It is freely available on with source code provided for all protocols, a protocol being a series of operations relevant for a given application, such as radiomic feature extraction, total metabolic tumor volume (TMTV) assessment, and image annotation (Figure 1). Extensive documentation, video tutorials, and efficient user support are offered. LIFEx reads DICOM (PET, SPECT, CT, MRI, US) and non-DICOM images. It supports spatial filters and voxel size resampling. Users can draw and manipulate volumes of interest (VOIs) using a graphical interface or load VOIs obtained using a third-party application. LIFEx can calculate a broad range of radiomic features in 2D and 3D in compliance with IBSI requirements and enables easy and traceable modification of the default calculation parameters, including voxel size, bin size, or bin number. Radiomic features can be computed at the regional level but also at the voxel level to produce radiomic maps (Escobar et al, Med Phys 2022). Moreover, recently introduced biomarkers, such as the normalized distance from hotspot to centroid (NHOC, Jiménez-Sánchez et al, PNAS 2021) reflecting tumor aggressiveness, are available. TMTV from PET/CT images can be easily calculated using a dedicated protocol, including a practical one-click interactive tool to add or remove any high-uptake region from the Maximum Intensity Projection views. When multiple lesions are present, various disease dissemination biomarkers can be automatically calculated. A module is proposed to quickly annotate images using predefined or customized menus for subsequent supervised machine learning. LIFEx can be used interactively, or operations can be scripted for repeated processing of many image series. All settings and results can be saved for traceability. Radiomic features extracted with the software can be easily analyzed using user-friendly statistical and machine learning software such as Orange (orangedatamining.com). LIFEx has been downloaded by over 6,300 users around the world and is cited in more than 500 publications.
Conclusions: LIFEx is a modular and upgradeable tool that evolves to meet the users’ needs. Being free and compatible with any operating system, it is widely adopted to conduct radiomic studies in a non-coding environment. By allowing researchers to reproduce previously published radiomic results, LIFEx is helping to significantly advance radiomics and AI studies.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.