Thursday 23 January 2025
Researchers have been exploring the potential of deep learning models in diagnosing Cushing’s syndrome, a rare hormonal disorder that can cause a range of symptoms including weight gain, high blood pressure, and excess hair growth. In a recent study, scientists compared the performance of various pre-trained models and foundational models in classifying facial images to identify individuals with Cushing’s syndrome.
The team found that Transformer-based models, such as Vision Transformer (ViT) and Swin Transformer, outperformed convolutional neural networks like ResNet and Densenet. The best-performing model was ViT-B-32, which achieved an F1 score of 85.71% in diagnosing Cushing’s syndrome using facial images.
The researchers also experimented with the freezing mechanism in the DINOv2 model, which significantly improved its performance. This suggests that the features learned through self-supervised learning in DINOv2 exhibit strong generalization capabilities.
Interestingly, the study revealed significant gender bias in the models’ performance. The DINOv2 model demonstrated a higher accuracy for female samples compared to male samples, with an F1 score of 75% for males and 84.62% for females. This disparity may be attributed to the imbalance in the number of male and female samples in the training dataset.
The team’s findings highlight the potential of deep learning models in diagnosing Cushing’s syndrome using facial images. However, they also underscore the need for more comprehensive approaches to address gender bias and ensure that models are robust and accurate across different populations.
In addition to its implications for medical diagnosis, this study has broader implications for the development of machine learning models. It suggests that Transformer-based architectures may be particularly well-suited for tasks requiring attention mechanisms, such as image classification.
The researchers’ use of facial images to diagnose Cushing’s syndrome is also noteworthy. Facial recognition technology has been increasingly used in various applications, from security systems to customer service chatbots. This study demonstrates the potential of this technology for medical diagnosis and highlights its potential for improving patient care.
As machine learning models continue to play an increasingly important role in healthcare, it will be essential to ensure that they are developed with fairness and accuracy in mind. The findings of this study serve as a reminder of the need for continued research into the development of robust and accurate machine learning algorithms, particularly in high-stakes applications like medical diagnosis.
Cite this article: “Deep Learning Models for Diagnosing Cushings Syndrome”, The Science Archive, 2025.
Cushing’S Syndrome, Deep Learning Models, Facial Images, Transformer-Based Models, Convolutional Neural Networks, Gender Bias, Machine Learning Algorithms, Medical Diagnosis, Image Classification, Attention Mechanisms
