ImageCLEF is an international initiative organized by the CLEF initiative lab, aimed at fostering the development of new AI and machine learning-based solutions to address challenges in the healthcare and medical domains. This year’s edition focused on developing the best possible solution for classifying tuberculosis types based on CT scan radiographic images.
Among synbrAIn’s many collaborations, the partnership with SenticLab is one of the most consistent and productive, especially in the healthcare sector. We are therefore delighted to share a recent achievement by SenticLab, which participated in this year’s ImageCLEFmed competition. Among all the research groups competing to develop the best solution for tuberculosis classification, SenticLab’s approach achieved the highest accuracy score!
The competition featured a total of 59 solutions, each designed to classify tuberculosis into five possible categories. The challenge was particularly complex due to subtle differences between some tuberculosis types, leading to a high risk of misclassification within the dataset. Additionally, many of the images provided in the competition dataset contained multiple types of tuberculosis, even though the competition required assigning a single label per image.
SenticLab’s solution employed a hierarchical approach. First, an initial classifier was trained exclusively on radiographic images containing a single tuberculosis type. This first model was designed to determine whether the input image contained any of the tuberculosis types considered in the competition. If a match was found, a second model (trained separately but on the same dataset) was then used to identify the most significant tuberculosis type in the image, incorporating the predictions from the first classification model.
This is not the first time SenticLab has won an ImageCLEF competition. In 2020, when the competition also focused on identifying lung diseases from radiographic images, SenticLab’s team emerged victorious, and their results were published in a scientific article featured in a Springer International journal. More details can be found in this post on our magazine.
SenticLab and synbrAIn continuously collaborate on developing AI-powered solutions for healthcare. Our partnership has allowed both teams to grow together, leading to the development of numerous machine learning approaches for medical data analysis, including the implementation of predictive models to support image-based diagnosis.
