Research status and progress of deep learning in automatic esophageal cancer detection

Abstract

Esophageal cancer (EC), a common malignant tumor of the digestive tract, requires early diagnosis and timely treatment to improve patient prognosis. Automated detection of EC using medical imaging has the potential to increase screening efficiency and diagnostic accuracy, thereby significantly improving long-term survival rates and the quality of life of patients. Recent advances in deep learning (DL), particularly convolutional neural networks, have demonstrated remarkable performance in medical imaging analysis. These techniques have shown significant progress in the automated identification of malignant tumors, quantitative analysis of lesions, and improvement in diagnostic accuracy and efficiency. This article comprehensively examines the research progress of DL in medical imaging for EC, covering various imaging modalities such as digital pathology, endoscopy, computed tomography, etc. It explores the clinical value and application prospects of DL in EC screening and diagnosis. Additionally, the article addresses several critical challenges that must be overcome for the clinical translation of DL techniques, including constructing high-quality datasets, promoting multimodal feature fusion, and optimizing artificial intelligence-clinical workflow integration. By providing a detailed overview of the current state of DL in EC imaging and highlighting the key challenges and future directions, this article aims to guide future research and facilitate the clinical implementation of DL technologies in EC management, ultimately contributing to better patient outcomes.

Keywords: Esophageal cancer; Artificial intelligence; Deep learning; Automatic detection; Medical imaging

Core Tip: Esophageal cancer (EC), a common malignant tumor, requires early detection for prognosis improvement. Deep learning (DL), particularly convolutional neural networks, has revolutionized EC diagnosis by enabling automated analysis of multimodal medical imaging, including digital pathology, endoscopy, and computed tomography. This article underscores the potential of DL to enhance screening accuracy and efficiency while addressing critical challenges such as constructing high-quality datasets, promoting multimodal feature fusion, validating model interpretability, and establishing dynamic evaluation systems. This article aims to guide future research and facilitate the clinical implementation of DL technologies in EC management.