Radiomics as a visual tool with potential clinical implications - A comment on the implementation of habitat radiomics in a liver cancer series

Sep 29, 2026

Uploaded by pop out

The correspondence discusses a study by Chen and colleagues proposing habitat radiomics for hepatocellular carcinoma, using CT-derived tumor segmentations to compute voxel-level Hounsfield Unit and entropy values. Tumor subregions, termed “radiomic habitats,” show strong links to survival and outperform standard whole-tumor radiomics in training and external validation. The letter highlights the role of spatial partitioning, pre-selecting informative intratumoral regions, and maps derived from local entropy to support “spatial radiomics” toward clinical implementation, while outlining key standardization and biological-mapping steps for future research.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages2

Reading Time5 min

AccessFree to download

Contents

Keywords

Letter to the Editors

Methods and Findings

Clinical Implications and Future Research

Additional Questions

Translational Oncology 64 (2026) 102640 Contents lists available at ScienceDirect Translational Oncology journal [homepage:](homepage: www.elsevier.com/locate/tranon)[ www.elsevier.com/locate/tranon](homepage: www.elsevier.com/locate/tranon) Radiomics as a visual tool with potential clinical implications: A comment on the implementation of habitat radiomics in a liver cancer series A R T I C L E I N F O Keywords: Radiomics Hepatocellular cancer Habitat Dear Editors, We read with great interest the paper by Chen K and colleagues, entitled “Habitat radiomics based on CT images to predict survival and immune status in hepatocellular carcinoma, a multi-cohort validation study,” published in Translational Oncology [1]. The authors analyzed CT images from a multicenter series of HCC patients and, after tumor segmentation, calculated voxel-by-voxel values of Hounsfield Unit (HU) and entropy. By the Otsu thresholding method, they identified tumor subregions, named “radiomic habitats,”having different combinations of the values (high or low) of the two indices (HU and entropy). The radiomic habitats had an excellent association with patient survival in training and external validation cohorts and outperformed standard radiomics. The present data are relevant as they enhance the clinical applicability of radiomics by pursuing an innovative approach. Several points of interest can be raised. Firstly, the analysis reported by Chen overcomes the concept of whole-tumor radiomics, which provides a single value for each index representing the entire neoplasm, representing a further step forward in leveraging imaging data. In a similar analysis conducted on 20 patients affected by colorectal metastases, we computed voxel-byvoxel tumor entropy and demonstrated that a quantitative analysis of intratumoral entropy values led to an accurate prediction of pathology data, outperforming the single entropy value provided by standard radiomics [2], precisely as Chen et al. observed in their HCC series. Secondly, the authors demonstrate that the analysis of an intratumoral region selected using a textural criterion outperforms the analysis of the entire mass. This concept may seem counterintuitive, as reducing the amount of information leads to improved model performance. However, it suggests that accurately pre-selecting regions with specific patterns is crucial for understanding the relationship between imaging features, tumor biology, and clinical outcomes. The same concept is applied when tumor biopsies are directed to the regions that are considered the most informative ones [3]. Finally, the most relevant concept is partitioning based on radiomic habitats, which reveals striking visual differences across patients and highlights intratumoral heterogeneity. This study contributes to a research approach described as “spatial radiomics.” This advancement is analogous to the transition from standard omics to spatial omics, such as spatial transcriptomics or proteomics, where pathology data are identified and precisely localized within the analyzed region [4–6]. The adoption of a spatial approach in radiomics seems almost selfevident: standard imaging inherently displays intratumoral heterogeneity, as demonstrated by Chen et al., who highlighted this through heterogeneous HU densities. So, why shouldn’t radiomics capitalize on this? Some studies in the literature, including ours, have explored this concept [2,7,8]. In our work, we constructed radiomic maps by encoding local entropy values into a palette cale, and computed voxel-by-voxel [2]. In this preliminary series, we identified intriguing entropy patterns that predicted tumor modifications following chemotherapy and appeared theoretically straightforward to correlate with clinical decisions (Fig. 1). For instance, the tumor profile on imaging plays a crucial role in planning appropriate surgery. Could the entropic profile serve as a reliable biomarker to guide surgical pla

FAQs

What does the study introduce as “radiomic habitats” in hepatocellular carcinoma?

It partitions tumors into subregions based on voxel-wise CT-derived Hounsfield Unit and entropy values, naming these subregions “radiomic habitats.”

How does the authors’ spatial partitioning approach compare with standard whole-tumor radiomics?

The letter states that habitat-based partitioning provides improved association with survival and outperforms standard radiomics by moving beyond a single whole-tumor value for each index.

What key steps does the letter recommend for future research to enable clinical implementation?

It calls for standardization and optimization of the technique (including reproducibility across kernels), clarification of the pathology/genetic basis for map regions, and possibly qualitative evaluation of pattern maps for easier clinical integration.

Sep 29, 2026

Uploaded by pop out