Leveraging large language models to analyze solid dosage disintegration images

Disintegration kinetics and behavior can affect the bioavailability and efficacy of solid dosage forms. Specifically, in vitro dissolution curves are commonly used to predict the in vivo behavior of solid dosage forms; however, they sometimes fail to provide insight into their formulation design and dissolution mechanisms. To overcome the limitations of standard dissolution tests, our group is developing a visual acquisition system to study the morphological changes of solid dosages under sink conditions without agitation via time-lapse macro-imaging. Recently, we introduced Dis-GPT, a large language model (LLM) designed for analyzing disintegration images of solid dosages. Trained on extensive biomedical and pharmaceutical literature, Dis-GPT automates and standardizes image analysis, generating precise descriptions and assessing disintegration similarities. Through repeated trials, Dis-GPT demonstrated consistent accuracy, aligning closely with the results of the hash algorithm. This innovation reduces manual analysis time, enhances reproducibility, and adapts to evolving research needs, promising broader applications in pharmaceutical research and beyond. This study introduces Dis-GPT, an LLM designed for analyzing solid dosage disintegration images under sink conditions without agitation. It automates and standardizes image analysis, enhancing accuracy and reproducibility for broader applications in pharmaceutical research.