Diagnostic performance of radiomics in adrenal masses: A systematic review and meta-analysis

Hao Zhang, Hanqi Lei, Jun PangFrontiers in Oncology 2022 · DOI

TL;DR

This systematic review and meta-analysis evaluates the diagnostic performance of radiomics in identifying adrenal masses, finding it promising for differentiating malignant from benign tumors. While radiomics shows potential, the study highlights significant heterogeneity and methodological limitations in current research, emphasizing the need for improved standardization and validation.

What Problem Does This Paper Address?

Adrenal incidentalomas, masses discovered incidentally during abdominal imaging, are increasingly common. While most are benign, a small percentage can be malignant. Accurately differentiating between benign and malignant adrenal masses is crucial for appropriate patient management, avoiding unnecessary biopsies or treatments for benign lesions, and ensuring timely intervention for malignant ones. However, conventional imaging techniques often struggle with this differentiation due to overlapping appearances and reliance on radiologist experience. Radiomics, a technique that extracts quantitative features from medical images, offers a potential solution by providing objective, data-driven insights. This paper addresses the need to systematically assess the current state of radiomics research in this area, evaluating its diagnostic performance and the quality of existing studies.

Methods

The researchers conducted a systematic review and meta-analysis of studies investigating radiomics for diagnosing adrenal masses. They searched multiple databases (PubMed, Embase, Web of Science, Cochrane Library) for eligible literature. To assess the quality and risk of bias of the included studies, they employed the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) and the Radiomics Quality Score (RQS). For the meta-analysis, they pooled data from studies that could provide a two-by-two contingency table to calculate key diagnostic performance metrics: pooled sensitivity, specificity, diagnostic odds ratio (DOR), positive likelihood ratio (PLR), and negative likelihood ratio (NLR). They also used the summary receiver operating characteristic (SROC) curve and area under the curve (AUC) to assess overall diagnostic accuracy. Heterogeneity was evaluated using the I² index, and subgroup analyses were performed to explore potential sources of variation, considering factors like CT type, feature type, and machine learning application. Publication bias was examined using Deeks' funnel plot.

Datasets

The study did not introduce new datasets. Instead, it analyzed data from 28 previously published studies that met their inclusion criteria. These studies utilized various imaging modalities, primarily CT (19 studies) and MRI (7 studies), with a few using FDG PET/CT. The "datasets" in this context are the collections of patient scans and associated diagnostic outcomes (histopathology or follow-up imaging) from these individual research papers.

Key Insights and Ideas

  • Radiomics shows promise, but it's not a magic bullet yet: The meta-analysis indicates that radiomics can achieve good diagnostic performance in differentiating malignant from benign adrenal tumors, with pooled sensitivity and specificity around 80%. This suggests it can be a valuable tool.
  • Heterogeneity is a major hurdle: A significant finding is the substantial heterogeneity across studies. This means that results can vary widely, making it difficult to draw definitive conclusions and hindering direct clinical translation.
  • CT-based radiomics is the most studied: The majority of included studies focused on CT imaging, highlighting its current dominance in this field.
  • Unenhanced and contrast-enhanced CT offer better performance: Subgroup analysis revealed that radiomics derived from both unenhanced and contrast-enhanced CT scans generally performed better than those using only contrast-enhanced CT.
  • Complexity can be a double-edged sword: While second-order or higher-order radiomics features can improve diagnostic sensitivity, they also appear to increase the rate of false positives. This suggests a trade-off between capturing subtle patterns and maintaining specificity.
  • Machine learning's role is still unclear: The study found no significant difference in diagnostic ability between studies that used machine learning algorithms and those that didn't. However, the limited number of studies employing machine learning means this conclusion should be interpreted cautiously.
  • Methodological quality needs improvement: A consistent theme is the suboptimal methodological quality and risk of bias in many studies, particularly concerning feature reduction, validation, and open science practices.

Key Results

  • Pooled Diagnostic Performance (CT-based radiomics):
    • Sensitivity: 0.80 (95% CI: 0.68-0.88)
    • Specificity: 0.83 (95% CI: 0.73-0.90)
    • AUC: 0.88 (95% CI: 0.85-0.91)
    • DOR: 19.06 (95% CI: 7.87-46.19)
  • Quality Assessment: The average Radiomics Quality Score (RQS) was low (5.11 out of a possible 36, representing 14.2% of the total score), indicating room for improvement in study design and reporting.
  • Risk of Bias: QUADAS-2 assessment indicated a moderate risk of bias across the included studies.
  • Heterogeneity: High heterogeneity was observed in both sensitivity (I² = 87.09%) and specificity (I² = 72.1%).
  • Subgroup Analysis: Radiomics based on unenhanced and contrast-enhanced CT showed higher DOR compared to contrast-enhanced CT alone. Studies using only first-order features had lower sensitivity but higher specificity than those combining with second-order or higher-order features.

Limitations and Future Work

The authors acknowledge several limitations:

  • Limited Literature: The meta-analysis was based on a relatively small number of studies, necessitating caution in interpreting the results.
  • Suboptimal Study Quality: The overall low RQS scores and moderate risk of bias in the included studies mean that the findings may be influenced by methodological weaknesses.
  • Significant Heterogeneity: The substantial heterogeneity across studies, potentially due to variations in pathological types, segmentation methods, reconstruction parameters, and feature extraction, makes it challenging to generalize findings.
  • Lack of Comparison with Radiologists: Most studies did not directly compare radiomics performance with that of experienced radiologists, limiting the assessment of its added clinical value.
  • Limited Validation: A significant portion of studies lacked external validation or prospective testing, which is crucial for confirming the generalizability and robustness of radiomics models.
  • No Grey Literature Included: The exclusion of grey literature might have introduced publication bias.

Future work should focus on:

  • Improving Methodological Rigor: Future studies need to adhere to standardized reporting guidelines, employ robust feature selection and validation strategies, and consider cost-effectiveness.
  • Standardization: Developing standardized radiomics workflows and protocols is essential to reduce heterogeneity and improve reproducibility.
  • Prospective and Multi-center Studies: Conducting prospective, multi-center studies with larger sample sizes and rigorous validation is critical for clinical translation.
  • Direct Comparison with Radiologists: Future research should include direct comparisons of radiomics performance against expert radiologists to better define its role in clinical decision-making.
  • Exploring Specific Histologic Types: More studies are needed to investigate radiomics performance for differentiating specific subtypes of adrenal masses.

Summary updated

Abstract

https://www.crd.york.ac.uk/PROSPERO/ CRD 42022331999 .
BibTeX
@inproceedings{zhang2022diagnostic,
  author = {Hao Zhang and Hanqi Lei and Jun Pang},
  title = {{Diagnostic performance of radiomics in adrenal masses: A systematic review and meta-analysis}},
  booktitle = {Frontiers in Oncology},
  year = {2022},
  doi = {10.3389/fonc.2022.975183},
}
Figure