This approach could potentially assist in the classification of cancers beyond retroperitoneal tumors. Professor Christina Messiou, MD, mentioned, “Our innovative technique leveraged disease-specific features, but with further refinement, this technology has the potential to improve outcomes for numerous patients each year.”
The Lancet Oncology published results from the retrospective, multi-cohort RADSARC-R study, revealing that an AI-based mammography model demonstrated efficacy in predicting the histological characteristics and diversity of retroperitoneal granulomas, offering valuable insights for risk stratification in this patient cohort.
An algorithm-based radiomic volume fraction (ARVF) model achieved an impressive area under the receiver operator curve (AUROC) of 0.928. In the phase 3 STRASS study (NCT01344018), which focused on histology classification in patients with retroperitoneal sarcoma, the model exhibited an accuracy rate of 0.843, a sensitivity of 0.9323, a specificity of 1.8929, and positive and negative predictive values of 0.480 and 0.984, respectively.
When applied to predict disease grade in the validation cohort, the radiomics model yielded an AUROC of 0.882. Additionally, investigators reported reliability, sensitivity, and specificity scores of 0.823, 0.865, and 0.865, with good predictive values of 0 and 0.778, respectively.
Lead author Amani Arthur, MRCPCH, a scientific research fellow at The Institute of Cancer Research, London, and affiliated with The Royal Marsden NHS Foundation Trust, emphasized the urgent necessity to improve the diagnosis and treatment of patients with retroperitoneal carcinoma, a population with historically poor outcomes.
Early studies indicated that leveraging imaging data led to the development of a powerful AI tool capable of rapidly and accurately identifying the type and grade of retroperitoneal melanomas compared to current methodologies. This advancement may expedite diagnosis, improve individual outcomes, and tailor treatments more effectively by precisely assessing each patient’s disease risk.
Incorporating data from a confirmation group involved in the phase 3 STRASS trial, which explored neoadjuvant therapy for retroperitoneal tumors, alongside insights from patients treated at Royal Marsden Hospital in London, researchers aimed to create radiomic classification models predicting morphological characteristics of retroperitoneal leiomyosarcoma and liposarcoma. The study utilized CT scan images from the discovery dataset to develop a radiomics workflow encompassing model construction, human delineation, sub-segmentation, feature extraction, and more.
Professor Christina Messiou, MD, a consultant radiologist at The Royal Marsden NHS Foundation Trust and a professor of Imaging for Personalised Oncology at the Institute of Cancer Research, London, expressed enthusiasm for the cutting-edge technology’s potential to enhance patient outcomes through expedited diagnosis and personalized treatment. The widespread adoption of this tool, especially in centers not specialized in retroperitoneal tumors, could ensure accurate disease identification and staging.
The study’s inclusion criteria encompassed patients aged 18 and above with confirmed leiomyosarcoma or retroperitoneal liposarcoma, emphasizing major unifocal conditions and artifact-free CT scan images depicting tumor volume accurately. Patient and record criteria in the confirmation cohort aligned with those in the discovery cohort.
The study involved 170 patients from the discovery cohort, with a median age of 63 years (range, 27 to 89), and 89 patients in the validation cohort, with a median age of 59 years (range, 33–77). Most patients in both cohorts exhibited ECOG or World Health Organization performance statuses of 0 (61% vs. 88%), liposarcoma (69%, 85%), and grade 2 disease (44%, 37%). Surgical intervention was the primary treatment modality for a higher percentage of patients in the discovery cohort (97%) compared to the validation cohort (48%).
Looking ahead, this methodology could potentially be applied to classify various types of malignancies beyond retroperitoneal sarcoma. Messiou concluded by highlighting the technology’s capacity to enhance prognoses for thousands of patients annually through ongoing algorithmic enhancements.
