AssistMED – researcher's assistant
AssistMED is a research assistant designed for the automatic analysis of medical data in cardiology. We have developed a platform to aid scientific research, along with three natural language processing (NLP) algorithms that accurately extract trustworthy information regarding diseases, medications and their dosages, as well as echocardiographic parameters from medical texts. This project was carried out in collaboration with the Medical University of Warsaw's scientific and research facility and academic center, funded through the "Innovation Incubator 4.0," where it emerged as one of the successful projects. A key requirement of the project was to achieve high operational precision while assessing the classification error rate. We also aimed to evaluate how well a medically knowledgeable individual performs in comparison to software when organizing data.
Our solution includes a customizable knowledge base, a proprietary fuzzy phrase search that's easy to modify, and integration with pre-trained language models (ML) tailored for the Polish language and MED7. Since our work involved Polish texts, we utilized the Google Translate API and established a translation cache. In our statistical reporting for various types of classified data, we compute metrics such as sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), effectiveness, Kappa coefficient, Fisher's coefficient, chi-squared, error matrix, Wilcoxon test, medians, and more. The architecture of our solution is designed for scalability and supports the integration of modules that can work with other systems.
We have achieved a remarkable effectiveness rate for the Polish language, ranging from 90% to 100%. This allows for the straightforward addition of new diseases and drugs and can also be applied in other areas of medicine. Potential applications include analyzing extensive medical data in cardiology, categorizing clinically relevant data, enhancing the efficiency of observational clinical trials, and enabling automatic risk assessments.