AI model shows accuracy in distinguishing mild/severe PDD from normal tissue

AI model shows accuracy in distinguishing mild/severe PDD from normal tissue

Researchers are currently working to improve the detection of thyroid eye disease in patients.

Reviewed by Paul S. Zhou, MD

According to Paul Zhou, a deep learning model for thyroid eye disease (TED) can improve disease detection and highlight the need or referral to oculoplastic surgeons and endocrinologists to ensure that patients diagnosed with disease receive treatment earlier to prevent permanent disability and loss of vision. MARYLAND.

Zhou is a researcher in the Department of Ophthalmic Plastic Surgery in the Department of Ophthalmology at Massachusetts Eye and Ear in Boston.

PDD is characterized by a wide range of symptoms, including dry eye, photophobia, diplopia, and decreased visual acuity and visual fields. If left untreated, the condition can be aesthetically devastating to patients, but can also increase the risk of compressive optic neuropathy.

Zhou and his colleagues seek to facilitate the detection of ASD through the use of an artificial intelligence (AI) model using X-ray imaging to screen patients and determine disease severity. Such a model would facilitate the diagnostic process and assessment of severity for physicians such as general practitioners, endocrinologists, and thyroid surgeons who may have less familiarity with PDD than general ophthalmologists and oculoplastic surgeons, a- he explained.

“Having such a tool that is readily available and helps in the diagnosis of PDD can be helpful,” Zhou said.

The AI ​​model

In the study, which may be the first to apply AI to TED in the United States, researchers set out to train an AI algorithm to accurately detect TED and identify compressive optic neuropathy.

“An AI model is an approximation of a real function that connects inputs and outputs,” he explained.

During the training process, the model can learn to spot subtle features, allowing the model to learn and relearn from the images it is exposed to.

In the 10-year study, researchers retrospectively looked at patients with orbital CT scans who had undergone examination by an oculoplastic surgeon. The dataset included patients with and without PDD.

A region of interest on the CT scans was selected and the left and right eyes were distinguished to allow independent AI training, he explained. The region of interest in the images was then classified as normal or as having mild or severe PDD based on clinical examination by an oculoplastic surgeon.

Datasets composed of normal orbits and mild and severe TED were transformed into color images, which resulted in more vibrant images in cases of severe TED surrounding the extraocular muscles and connective tissues.

Investigators used VGG16 (also called OxfordNet), a convolutional neural network model named after the Visual Geometry Group at Oxford University. VGG16 is 16 layers deep and was previously trained on ImageNet, which contains over a million images. When the region of interest in the CT images was introduced into VGG16, the model was able to differentiate between normal thyroid, mild TED and severe TED.

In this study, 885 images of 131 patients were used, of which 279 were normal, 251 showed mild TED and 355 showed severe TED; 100 images of the total were retained for further evaluation.

Zhou noted that the overall prediction accuracy across the 3 groups was 94.27%. Normal and mild PDD cases were never misclassified as severe PDD. Of the 355 cases of severe PDD, 1 was misclassified as mild disease.

This model can help general practitioners distinguish between normal thyroid tissue and that with mild PDD. “The AI ​​model can make this distinction based on a snapshot, with an accuracy of 92.16%,” Zhou said.

When the AI ​​model was tested using the 100 retained images for later evaluation, the accuracy was 98%. Another test pitted the AI ​​model against a doctor: When 114 randomly selected unlabeled images were scored by an oculoplastic surgeon, the surgeon’s accuracy was 43.83%.

Future efforts will incorporate the model into radiology protocols, compare the model’s accuracy with other human experts, and apply similar machine learning to other eye conditions such as orbital tumors and inflammation.

Paul Zhou, MD


Zhou is a researcher at Massachusetts Eye and Ear in Boston and has no financial interest in this subject.

#model #shows #accuracy #distinguishing #mildsevere #PDD #normal #tissue

Leave a Comment

Your email address will not be published. Required fields are marked *