Multimodal artificial intelligence system for detecting a small esophageal high-grade squamous intraepithelial neoplasia: A case report

Abstract

BACKGROUND

Recent advancements in artificial intelligence (AI) have significantly enhanced the capabilities of endoscopic-assisted diagnosis for gastrointestinal diseases. AI has shown great promise in clinical practice, particularly for diagnostic support, offering real-time insights into complex conditions such as esophageal squamous cell carcinoma.

CASE SUMMARY

In this study, we introduce a multimodal AI system that successfully identified and delineated a small and flat carcinoma during esophagogastroduodenoscopy, highlighting its potential for early detection of malignancies. The lesion was confirmed as high-grade squamous intraepithelial neoplasia, with pathology results supporting the AI system’s accuracy. The multimodal AI system offers an integrated solution that provides real-time, accurate diagnostic information directly within the endoscopic device interface, allowing for single-monitor use without disrupting endoscopist’s workflow.

CONCLUSION

This work underscores the transformative potential of AI to enhance endoscopic diagnosis by enabling earlier, more accurate interventions.

Key Words: Artificial intelligence; Multimodal artificial intelligence system; Esophageal squamous cell carcinoma; High-grade intraepithelial neoplasia; Case report

Core Tip: This study introduces a novel multimodal artificial intelligence system (MAIS) based on the QueryInst network for real-time detection and delineation of esophageal squamous cell carcinoma and precancerous lesions during endoscopy. Unlike traditional artificial intelligence systems, MAIS integrates directly into the endoscopic device, allowing for single-monitor use without altering the endoscopist’s workflow. This case report demonstrates its ability to accurately identify a flat esophageal lesion, which was confirmed as high-grade squamous intraepithelial neoplasia. The findings highlight potential of MAIS for improving early diagnosis and biopsy accuracy in high-risk gastrointestinal conditions such as esophageal squamous cell carcinoma.

INTRODUCTION

Artificial intelligence (AI) has developed rapidly in recent years in terms of endoscopic-assisted diagnosis of gastrointestinal diseases[1]. Several clinical studies have reported the auxiliary diagnostic performance of AI in clinical practice, highlighting the great potential of its application in real clinical settings. Considering the high morbidity and medical burden of esophageal squamous cell carcinoma (ESCC) worldwide, our team has successfully developed a multimodal AI system (MAIS) based on the QueryInst network, which can detect and delineate ESCC and precancerous lesions in real-time for accurate biopsy and early diagnosis[2,3]. Unlike other AI systems that required an additional monitor, MAIS was integrated directly into the endoscopic device that endoscopists used daily, i.e., single-monitor use, without changing the endoscopists’ operating habits.

CASE PRESENTATION

Chief complaints

A 47-year-old man was found to have a small flat mucosal esophagus lesion during esophagogastroduodenoscopy assisted by MAIS.

History of present illness

The patient underwent esophagogastroduodenoscopy in our hospital a health examination. A small and flat mucosal lesion in the esophagus was revealed by the esophagogastroduodenoscopy.

History of past illness

The patient had no history of past illness.

Personal and family history

The patient denied any family history of malignant tumors.

Physical examination

His vital signs were as follows: Body temperature, 37.0 °C; blood pressure, 101/61 mmHg; heart rate, 68 beats per minute; respiratory rate, 15 breaths per minute. The patient also had clear breath sounds bilaterally. A soft, non-tender, abdomen, bowel sounds, and no hepatomegaly or splenomegaly.

Laboratory examinations

Laboratory examinations were absent.

Imaging examinations

During upper endoscopy, MAIS identified a flat esophageal mucosal lesion approximately 0.5 cm in diameter 31 cm from the incisors under white-light imaging, narrow band imaging (NBI), magnified NBI, and iodine staining (Figure 1A-E). The flat lesion was slightly red under white light endoscopy and brown under NBI. After magnification, the intrapapillary capillary loop was B1 type by AB classification for ESCC and lightly stained after iodine staining. MAIS not only detected and delineated the lesion but also showed the current endoscopic imaging modality with the probability value of the lesion being a cancerous lesion in the upper left of the endoscopic monitor.

Figure 1 The multimodal artificial intelligence system identified a small and flat esophageal mucosal lesion of approximately 0.5 cm in diameter under four endoscopic imaging modalities. A: White-light imaging (blue dashed line); B: Narrow band imaging (yellow dashed line); C: Magnified narrow band imaging (yellow dashed line); D: Iodine staining (blue dashed line); E: Application of the multimodal artificial intelligence system; F: Histopathology of the resected specimen showed the lesion was a high-grade squamous intraepithelial neoplasia (hematoxylin and eosin, × 200).

FINAL DIAGNOSIS

Histopathology of the specimen resected by endoscopic submucosal dissection demonstrated disordered cell polarity and nuclear atypia and enlargement, and the lesion was confirmed to be a high-grade squamous intraepithelial neoplasia measuring 3.0 mm × 2.0 mm (Figure 1F).

TREATMENT

The esophagus lesion was removed by endoscopic submucosal dissection.

OUTCOME AND FOLLOW-UP

The lesion was curatively resected, and follow-up endoscopy was planned in 6 months.

DISCUSSION

It was reported that MAIS was used to detect a small flat-type ESCC and a hypopharyngeal precancerous lesion incidentally, and showed great screening potential. We here report the smallest esophageal precancerous lesion detected by MAIS in real-time during clinical endoscopy, which was confirmed to be high-grade squamous intraepithelial neoplasia. Some large prospective studies also showed that AI improved the detection rate and miss rates of early ESCC and precancerous lesions during endoscopy. However, it is challenging to detect and delineate small flat precancerous lesions in real-time with AI, which can be missed even by senior endoscopists without AI (with experience of ≥ 10000 endoscopies)[4]. If the missed lesions developed into cancer insidiously, they would lead to heavy and economic burdens for families and society[5]. The biggest advantage of MAIS was that these four imaging modalities could meet the actual needs of hospitals different level and endoscopists to facilitate easier promotion and application of the system, especially in areas with limited medical resources. However, consistent false detection by MAIS is an unavoidable problem, which might cause anxiety among endoscopists and unnecessary biopsies. We will continue optimizing the model performance and believed that MAIS would assist endoscopists in detecting more early-stage ESCCs in the future, improving patients’ prognosis.

CONCLUSION

This case highlights the significant potential of the MAIS in enhancing the detection and diagnosis of early ESCC and precancerous lesions during endoscopy. As demonstrated in this case, high-grade squamous intraepithelial neoplasia can be in real-time, and MAIS offers a reliable and integrative solution for endoscopists’ daily practices. Its ability to operate seamlessly across various imaging modalities and identify tiny lesions, often challenging even for experienced clinicians, underscores its utility in advancing early cancer detection. Despite some limitations, such as false detections, the system has the advantages in terms of accessibility and adaptability, particularly for resource-constrained settings, which make it a promising tool for broader implementation. Continued optimization of MAIS could further elevate its diagnostic accuracy, ultimately reducing the clinical and economic burden of ESCC through earlier intervention and improved prognoses. This case reaffirms the transformative role of AI-driven systems in modern medical diagnostics.