Radiomics has shown promise in some studies of patients with endometrial cancer, but there is “insufficient evidence” to determine if radiomics provides a clinical benefit, according to a review published in the European Journal of Surgical Oncology.1

The review included 17 studies testing radiomics — the use of artificial intelligence (AI) to analyze radiographic imaging — for preoperative assessment of endometrial cancer.

The radiomics approaches tested performed well in predicting cancer severity, but most of the studies did not meet quality standards, according to the review authors. The authors therefore concluded that more studies are needed to demonstrate a benefit in clinical decision-making.

Continue Reading

“The 2 main outcomes [of the review] are that we can probably anticipate the tumor differentiation with high probability with only noninvasive imaging with this technology, and we probably can also anticipate the risk of lymphadenopathy involvement by doing this kind of analysis,” said review author Benoît Gallix, MD, PhD, chief executive officer of the Institute of Image-Guided Surgery in Strasbourg, France.

“Most of the studies [that did not meet quality standards] are probably good, but the methodology and design are not strong enough,” Dr Gallix added, pointing out that he helped lead one of the studies that was excluded from the review due to quality issues.2

The main problem with the lower-quality studies, Dr Gallix said, was that they did not validate the AI models in a second group of patients and in a different setting or by using a different piece of imaging equipment. However, this type of validation has now become standard, he noted.

Review Details

The review began with 17 studies that applied AI to interpret radiographic imaging for staging endometrial cancer based on a range of characteristics.

To analyze the quality of the studies, the researchers developed a scoring system called the Simplified and Reproducible AI Quality Score (SRQS). This system is easier to use and does a better job of assessing clinical relevance than previous scoring systems, according to Dr Gallix.

Ten of the 17 studies scored below the quality cutoff (SRQS score greater than 10/20), leaving 7 studies that met quality standards.

Dr Gallix and colleagues noted that the 7 studies “demonstrated strong diagnostic performance in various objectives.” The objectives included myometrial invasion (n=2), lympho-vascular space invasion (n=1), lymph node involvement (n=2), high-risk endometrial cancer (n=1), and differential diagnosis between an endometrial precancerous lesion and early-stage carcinoma (n=1).

One of the studies included 530 patients and was designed to evaluate the diagnostic performance of a deep learning model in assessing myometrial invasion.3 The model proved more accurate than radiologist assessment (84.8% vs 78.3%), but accuracy was highest when AI and radiologists were used in tandem (86.2%).

In a study of 144 patients, researchers evaluated lympho-vascular space invasion in preoperative imaging.4 They developed a nomogram based on clinical features and a radiomics score. The nomogram demonstrated a sensitivity of 94%, a specificity of 78.6%, and an area under the curve (AUC) of 0.807.

In a study of 717 patients, researchers developed a nomogram based on clinical and radiomics data to predict preoperatively high-risk endometrial cancers and guide surgical management.5 The nomogram was effective for predicting high-risk cancers (AUC between 0.896 and 0.919) and for correcting clinical decisions.

Dr Gallix and colleagues noted, however, that even these higher-quality studies had limitations, and none of the studies provided a tool with short-term applicability in clinical practice.