As part of MetaOptima’s initiative to develop and implement intelligent dermatology software and tools designed to support medical professionals in their daily practice, the company’s Research & Development team has published the following research papers. For any questions relating to MetaOptima’s research please contact us at email@example.com
Diagnostic Accuracy Of Content-Based Dermatoscopic Image Retrieval With Deep Classification Features
The utilization of dermoscopic images for the study and diagnosis of skin-related diseases is a essential resource for the early detection of life-threatening skin diseases such as melanoma. Artificial intelligence (AI) is a powerful tool to assist in the classification of skin images with the use of convolutional neural networks (CNNs) while providing a high degree of accuracy to assist medical professionals in the identification and diagnosis of skin lesions. Visual features learned by a CNN, known as “deep features”, play a key role in identifying visually similar dermatoscopic images.
This work focuses on comparing the diagnostic accuracy of CNN trained for classification against results obtained from a more interpretable content-based image retrieval (CBIR) method. This involves querying a target image against a dataset of previously identified lesions. CBIR was tested using dermatoscopic images from three different datasets, and while the power of CNNs has been extensively used in the classification of skin imaging, results show that a CBIR-based method can reach similar performance to that of a classification network. However, this approach offers the added value of enabling CNNs to recognize unknown disease classes in new datasets, thus helping to improve diagnostic accuracy in a routine clinical setting.
Full article available at: https://onlinelibrary.wiley.com/doi/epdf/10.1111/bjd.17189
Multimodal Skin Lesion Classification Using Deep Learning
AI technology has allowed for the accurate identification of skin lesions through the implementation of convolutional neural networks (CNNs). Traditionally, systems render a diagnosis based on a single macroscopic image, the following study utilizes multiple imaging modalities with the combination of patient metadata in a multi-analytic approach with the goal of improving the performance of automated skin lesion diagnosis. Additionally, the system was tested in a five class classification task in order to better resemble a real-life clinical scenario. Results demonstrate the system outperforms both the baseline classifier utilizing a single macroscopic image when performing multiclass classification.
Full article available at: https://onlinelibrary.wiley.com/doi/epdf/10.1111/exd.13777
To learn more about MetaOptima’s initiatives in the artificial intelligence for dermatology research space we encourage you to read the following articles:
-The MetaOptima Team
Topics: Dermatology Visual Search CBIR Artificial Intelligence Content Based Image Retrieval Big Data Artificial Intelligence in Dermatology Intelligent Dermatology Software Big Data in Healthcare Intelligent Dermatology