Post written by Mingyan Cai, MD, PhD, FASGE, from the Endoscopy Center and Endoscopy Research Institute, Zhongshan Hospital, Fudan University, Shanghai, China.

In this study, we developed and validated an automatically optimized radiomics modeling system (AORMS) based on EUS images to diagnose small GI stromal tumors (GISTs) (<2 cm) from non-GISTs and predict those with high malignancy potential.

Because of the lack of effective diagnostic tools, endoscopists face a dilemma when dealing with small submucosal tumors (SMTs) (<2 cm). Most small SMTs are considered low risk; however, cumulating evidence has shown the potential malignancy of small SMTs should not be ignored and might be underestimated. EUS was considered the first-choice imaging method for the evaluation of SMTs.

Yet, relying solely on EUS yields a diagnostic rate of <50%, especially for small SMTs. EUS-guided fine-needle biopsy (FNB) has substantially augmented the accuracy of SMT diagnosis, but the universal implementation of FNB for all small SMTs could introduce additional medical costs, particularly in resource-constrained settings.

Moreover, accessibility of FNB is uneven in the worldwide setting. Given these considerations, a critical need persists for a reliable diagnostic method for the identification of small GISTs (<2 cm), especially for high-risk GISTs.

Currently, there is no clear consensus on the management of small gastric SMTs (<2 cm). AORMS can assist in clinical decision making, compensating for the limits of EUS assessment of small SMTs and allowing patients to receive timely and appropriate management. The article presents our proposed algorithm in a real-time AORMS setting, which is the next step of development.

After AORMS evaluation, regular follow-up is recommended for small SMTs, if excluding the possibility of GISTs. For a suspicious high-risk GIST, endoscopic resection or at least FNB is necessary. For a suspicious low-risk GIST, FNB or endoscopic resection will depend on further comprehensive evaluation, including clinical risk factors (such as GIST or other malignant tumor history and related family history), previous clinical examinations (including endoscopy, other imaging examinations, and tumor biomarkers), and the patient’s preferences.

For patients with related clinical high-risk factors, FNB or endoscopic resection will be recommended. On the contrary, patients without these factors are advised to follow up regularly.

AORMS shows good diagnostic performance and risk stratification capabilities for small SMTs (<2 cm). These promising results provide a foundation for further refining AORMS and integrating it with EUS equipment. This integration will enable endoscopists to perform onsite and real-time evaluations of small SMTs during endoscopic examinations, thereby contributing to improved management and decision making regarding small SMTs (<2 cm).

Overview of the experimental workflow. The representative EUS images of each patient were selected in which the lesions were manually segmented and delineated as the region of interest (left). Diagram of the AORMS framework (middle). Evaluation of the diagnostic performance of AORMS (right). AORMS, Automatically optimized radiomics modeling system; LR+, positive likelihood ratio; LR–, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value; ROC, receiver-operating characteristic.

Read the full article online.

The information presented in Endoscopedia reflects the opinions of the authors and does not represent the position of the American Society for Gastrointestinal Endoscopy (ASGE). ASGE expressly disclaims any warranties or guarantees, expressed or implied, and is not liable for damages of any kind in connection with the material, information, or procedures set forth.