Improving utility of brain tumor confocal laser endomicroscopy: objective value assessment and diagnostic frame detection with convolutional neural networks

TL;DR

This study develops a deep learning model using AlexNet to automatically classify brain tumor confocal laser endomicroscopy images as diagnostic or non-diagnostic, achieving high accuracy and reliability to assist during surgery.

Contribution

The paper introduces a deep learning approach with AlexNet for real-time classification of CLE images, improving efficiency in brain tumor surgery diagnostics.

Findings

Achieved 91% overall accuracy in classifying images.

High sensitivity and specificity (~91%) in detection.

ROC AUC of 0.958 indicating excellent model performance.

Abstract

Confocal laser endomicroscopy (CLE), although capable of obtaining images at cellular resolution during surgery of brain tumors in real time, creates as many non-diagnostic as diagnostic images. Non-useful images are often distorted due to relative motion between probe and brain or blood artifacts. Many images, however, simply lack diagnostic features immediately informative to the physician. Examining all the hundreds or thousands of images from a single case to discriminate diagnostic images from nondiagnostic ones can be tedious. Providing a real-time diagnostic value assessment of images (fast enough to be used during the surgical acquisition process and accurate enough for the pathologist to rely on) to automatically detect diagnostic frames would streamline the analysis of images and filter useful images for the pathologist/surgeon. We sought to automatically classify images as diagnostic or non-diagnostic. AlexNet, a deep-learning architecture, was used in a 4-fold cross validation manner. Our dataset includes 16,795 images (8572 nondiagnostic and 8223 diagnostic) from 74 CLE-aided brain tumor surgery patients. The ground truth for all the images is provided by the pathologist. Average model accuracy on test data was 91% overall (90.79 % accuracy, 90.94 % sensitivity and 90.87 % specificity). To evaluate the model reliability we also performed receiver operating characteristic (ROC) analysis yielding 0.958 average for the area under ROC curve (AUC). These results demonstrate that a deeply trained AlexNet network can achieve a model that reliably and quickly recognizes diagnostic CLE images.