Internally, we continuously retrain the model with additional data, provide updates to the architecture, and validate the model. Once the model has been validated internally, we periodically deploy it.

Absolutely. If the requested organ is deemed critical for radiation therapy in a general context, we can prioritize it in our internal labeling tasks and provide updates accordingly. However, updating the AI model requires hundreds of data samples, so the process may take several months. Collaboration in research can expedite this process, and we welcome any partnership requests!

In the past, we mainly utilized 3D models. However, 2D models offer advantages in detail resolution due to the absence of GPU memory constraints, although they may encounter connectivity issues between slices. On the other hand, 3D models often excel in connectivity but may have limitations in detail resolution. In the new development version, we aim to enhance the efficiency of AI model inference, leveraging the strengths of both 2D and 3D models while mitigating their weaknesses.

Currently, we provide over 100 normal organs and clinical target volumes (CTV) based on CT imaging. By 2025, we aim to expand our offerings to include over 200 structures. Additionally, we are developing AI models for various imaging techniques such as MRI, Cone-beam CT (CBCT), and Megavoltage CT (MVCT), which will be distributed to users through license renewals in the near future. In the long term, we are also working on providing automatic segmentation information for Gross target volume (GTV)!

Yes, we are currently developing a new version of OncoStudio, and we plan to introduce a cloud-based product in the first half of 2025.