AI Model May Reliably Predict Disease Histology in Retroperitoneal Sarcoma

“In the future, this approach may help characterize other types of cancer, not just retroperitoneal sarcoma. Our novel approach used features specific to this disease, but by refining the algorithm, this technology could one day improve the outcomes of thousands of patients each year,” according to Professor Christina Messiou, MD.

An artificial intelligence (AI)–based radiomics model demonstrated reliability when predicting the histological grade and type of retroperitoneal sarcomas, which may help with risk stratification in this patient population, according to findings from the retrospective, multi-cohort RADSARC-R study published in Lancet Oncology.¹

A radiomics plus algorithm-based radiomic volume fraction (ARVF) model produced an area under the receiver operator curve (AUROC) of 0.928 (P <.0001) for histology classification among a validation cohort of patients with retroperitoneal sarcoma in the phase 3 STRASS study (NCT01344018). Additionally, this model demonstrated an accuracy score of 0.843, a sensitivity of 0.923, a specificity of 0.829, a positive predictive value of 0.480, and a negative predictive value of 0.984.

The radiomics model, when employed for predicting disease grade in the validation cohort, elicited an AUROC of 0.882 (P <.0001). Investigators also reported scores of 0.823 for accuracy, 0.800 for sensitivity, 0.848 for specificity, 0.865 for positive predictive value, and 0.778 for negative predictive value.

“There is an urgent need to improve the diagnosis and treatment of patients with retroperitoneal sarcoma, who currently have poor outcomes,” lead author Amani Arthur, MRCPCH, clinical research fellow at The Institute of Cancer Research, London, and registrar at The Royal Marsden NHS Foundation Trust, said in a press release.²

“Through this early research, we’ve developed an innovative AI tool using imaging data that could help us more accurately and quickly identify the type and grade of retroperitoneal sarcomas than current methods. This could improve patient outcomes by helping to speed up diagnosis of the disease, and better tailor treatment by reliably identifying the risk of each patient’s disease.”

Investigators of this study assessed data from patients with retroperitoneal sarcoma in a discovery cohort treated at Royal Marsden Hospital in London, as well as data from a validation cohort of patients enrolled on the phase 3 STRASS study, which evaluated neoadjuvant radiotherapy in retroperitoneal sarcoma.¹ The study’s primary objective was developing radiomic classification models that could forecast the histological type and grade of retroperitoneal leiomyosarcoma and liposarcoma. CT scan images in the discovery dataset led to the development of a radiomics workflow including features such as manual delineation, sub-segmentation, feature extraction, and predictive model building.

“We’re incredibly excited by the potential of this state-of-the-art technology, which could lead to patients having better outcomes through faster diagnosis and more effectively personalized treatment,” senior author Professor Christina Messiou, MD, consultant radiologist at The Royal Marsden NHS Foundation Trust and a professor in Imaging for Personalised Oncology at The Institute of Cancer Research, London, said in a press release.² “As patients with retroperitoneal sarcoma are routinely scanned with CT, we hope this tool will eventually be used globally, ensuring that not just specialist centers—who see patients [with sarcoma] every day—can reliably identify and grade the disease.”

Patients 18 years and older with histologically confirmed retroperitoneal liposarcoma or leiomyosarcoma were eligible to be included in the study’s discovery cohort.¹ Additional eligibility requirements included having primary and unifocal disease and CT scan images that entirely captured the tumor volume without artifacts. Patient and scan criteria for the validation cohort were similar to those used for the discovery cohort.

Overall, the study included 170 patients in the discovery cohort, who had a median age of 63 years (range, 27-89). Additionally, 89 patients made up the validation cohort, who had a median age of 59 years (range, 33-77). Most patients in the discovery and validation cohorts, respectively, had an ECOG or World Health Organization performance status of 0 (61% vs 88%), liposarcoma (69% vs 85%), and grade 2 disease (44% vs 37%). A higher proportion of patients in the discovery cohort were treated with surgery only (97%) compared with those in the validation cohort (48%).

“In the future, this approach may help characterize other types of cancer, not just retroperitoneal sarcoma. Our novel approach used features specific to this disease, but by refining the algorithm, this technology could one day improve the outcomes of thousands of patients each year,” Messiou concluded.²

References

1. Arthur A, Orton MR, Emsley R, et al. A CT-based radiomics classification model for the prediction of histological type and tumour grade in retroperitoneal sarcoma (RADSARC-R): a retrospective multicohort analysis. Lancet Oncol. 2023;24:1277-1286. doi:10.1016/S1470-2045(23)00462-X
2. AI twice as accurate as a biopsy at grading aggressiveness of some sarcomas. News release. The Institute of Cancer Research. November 1, 2023. Accessed November 17, 2023. https://shorturl.at/ghopB