A Comprehensive Literature Review on Skin Disease Identification Through Deep Learning-Based Bi-Directional Image-Text Matching Techniques

Over the past few years, matching visual and textual data commonly known as image text matching has become increasingly important due to the rising demand for multimodal data understanding. ITM involves retrieving an image that matches a textual description, or vice versa, by evaluating the semantic similarity between visual and textual inputs. This task becomes particularly critical in medical applications such as skin disease identification, where accurate matching between clinical images and descriptive diagnostic text can aid in early detection and decision-making. One of the core challenges in this domain lies in bridging the semantic gap between visual features of skin lesions and medical terminology used in clinical notes or diagnostic reports. Despite various deep learning approaches developed for general ITM tasks, limited research has focused on domain-specific applications like dermatology. This paper provides a focused review of deep learning-based ITM frameworks, emphasizing their relevance to skin disease analysis. We explore model components such as image and text encoders, common embedding strategies (e.g., global, local, hybrid), and learning paradigms including adversarial, ranking, and pre-trained vision-language models. Additionally, benchmark datasets are discussed, including synthetic and real-world skin image–text pairs. The paper concludes by identifying key challenges such as dataset scarcity, domain adaptation, and explainability highlighting opportunities for advancing ITM in dermatological diagnostics.