Brought to you by

Letter to the Editor: “Current status and quality of radiomics studies in lymphoma: a systematic review”

by Martina Sollini, Margarita Kirienko, Arturo Chiti (martina.sollini@cancercenter.humanitas.it; martinasollini@msn.com )

Current status and quality of radiomics studies in lymphoma: a systematic review

Dear Editor,

We read with great interest the article by Wang et al “Current Status and Quality of Radiomics Studies in Lymphoma: A Systematic Review” [1]. The paper aims at providing an overview of the state-of-the-art and quality of radiomic studies in lymphoma. The Authors selected 45 papers on the field, 30 focused on diagnosis and classification; 16 on treatment response and outcome prediction (one study exploring both diagnostic and prognostic role of radiomics in non-Hodgkin’s lymphoma, was calculated twice). Studies were assessed according to different scores from the “classical” QUADAS-2 [2], to the “technical” radiomics quality score (RQS) [3], and the “clinical” phase analysis [4].

Radiomics has certainly proved to be one of the main domain of interest in image analysis, as shown by the number of publications in the field. A search in PubMed/Medline with the terms ((“texture analysis”) OR (“textural analysis”) OR (“textural features”) OR (“radiomic”) OR (“radiomics”)) resulted in more than 5340 papers (more than 3180 in the period 2018-2020). The same search in Embase resulted in 8044 records (approximately 4090 in the period 2018-2020). In the era of personalized medicine, potentially more accurate methods for the biological characterization of a certain disease, beyond “classical” risk factors and stage, are doubtless attractive. Moreover, radiomics relying on quantification is theoretically objective, reproducible, comparable, and potentially more informative than human eyes. Therefore, interest in radiomics is not surprising. Nonetheless, growing evidence pointed out the remarkable immaturity of the discipline, in terms of scarce quality of the studies which are burdened by clinical, technical, or methodological biases. Surprisingly, radiomic studies in lymphoma are even poorer than in solid tumours.

Wang et al. [1] classified at high/unclear risk of bias an increasing percentage of papers ranging from 33% up to 80%, based on the considered domain (from index test, to reference standard, to patient selection, and to flow and timing). The mean RQS was 14%, lower than 10% in more than half of the studies. None of the studies was classified as I, III, or IV. The majority of the studies were classified as discovery science or phase 0 (71% and 27%, respectively); only one study was classified as phase II (IIb according the updated version of classification [5]). The main concerns on radiomic studies in lymphoma were related to lack of validation, small sample size or retrospective design, as well as lack of a rigorous procedure. It would have been interesting also to evaluate the interdisciplinarity of the analyzed studies, since in solid tumour the composition of the research team has been suggested as one of the essential requirements to ensure advancement of radiomics toward clinical practice [5].

As discussed by Wang et al., the main challenges in lymphoma radiomic analysis include target lesion selection and segmentation. Tumour volume significantly influences textural features, but many papers on lymphoma did not report the size of the lesion, nor the criterion for the target lesion selection. Moreover, as well-known, changing the segmentation method, the region/volume of interest differs and ultimately affects textural features. However, the peri-tumour area seems to contain meaningful information, potentially contributing to distinguish brain tumour histology [6]. Nonetheless, in other anatomical districts, the inclusion of non-tumour area certainly includes noise, being many different tissues typically involved by lymphomatous lesions, potentially different in terms of contrast and signal-to-noise ratio of images. In addition to these items, we recently found that a random target lesion selection should not be adopted for radiomics applications in lymphoma unless the informative content of the lesion is demonstrated to be the same in terms of heterogeneity.

In our cohort, intra-patients lesions’ similarity differed between nodal and extra-nodal lesions and between long-term responders and refractory/relapsing HL [7]. Moreover, we proposed a novel feature selection approach based on volume-criteria. Accordingly, features were previously selected if they were highly correlated or uncorrelated to the volume. The rationale for volume-related criteria is to identify all potentially relevant information and discard collinear variables, without ignoring volume component that may be relevant to predict disease aggressiveness.

Collectively, the methods for target lesions selection and segmentation are crucial issues in radiomic methodology, regardless of radiomic analysis focused on solid tumour or haematological malignancy. However, lymphoma is a systemic disease typically characterized by the coexistence of multiple lesions with variable size, which might involve different tissues/organs. Accordingly, as confirmed by Wang et al., weak methodologies have a considerable impact on radiomic studies, and especially in lymphoma, this uncertainty results in unreliable findings. Radiomic trial should be rigorous to achieve robust and reproducible results and be implemented in clinical practice.