ESOPRED

Detection of Esophageal Cancer Subtypes Using High-Resolution Histopathology Images

Issue

Esophageal cancer (EC) is a serious global health issue, ranked as the 7th most common cancer and the 6th leading cause of cancer deaths worldwide. With over 600,000 new cases annually, the prognosis for patients is often poor, especially due to the difficulty in detecting the disease early. By the time symptoms such as difficulty swallowing or chest pain appear, the cancer is frequently in advanced stages. The 5-year survival rate remains below 20%, largely because over 70% of patients are diagnosed at an advanced stage, where treatment options are limited. One of the biggest challenges is accurately diagnosing EC, particularly its two main types: adenocarcinoma and squamous cell carcinoma. Current methods are often slow, subjective, and error-prone, making it harder for doctors to provide fast and accurate diagnoses. There is an urgent need for better tools to identify and classify this cancer early, so patients can get the right treatment on time.

Approach

An automated tool has been developed to quickly and accurately identify the type of esophageal cancer using tissue images. The tool analyzes high-quality images of tissue samples and uses artificial intelligence to classify the cancer as either adenocarcinoma or squamous cell carcinoma. This allows doctors to make faster, more reliable diagnoses. With just a simple upload of tissue images, the tool can instantly provide clear results, helping doctors make better decisions and reduce diagnostic errors. (The Technical Stuff: To build and test the model, the high-resolution histopathology images from The Cancer Genome Atlas (TCGA) Esophageal Carcinoma dataset were segmented into patches and analyzed using a pre-trained ResNet101 model for feature extraction. The extracted features were then evaluated using five machine learning classifiers, with the Feed-Forward Neural Network (FFNN) achieving the best performance, followed by Support Vector Classifier (SVC) and Logistic Regression (LR).)

Result

This AI-powered tool has shown a 94% accuracy rate in identifying esophageal cancer types (with an AUC ‘Area under the Curve’ score of 0.92 with the FFNN model). It helps doctors diagnose cancer faster and more accurately, especially in places where resources are limited. By supporting early detection, the tool increases the chances of successful treatment, improves patient outcomes, and helps reduce the number of lives lost to this aggressive cancer.