Julia Schnabel, AI-enabled imaging
Top builders in biotech
Axial: https://linktr.ee/axialxyz
Axial partners with great founders and inventors. We invest in early-stage life sciences companies such as Appia Bio, Seranova Bio, Delix Therapeutics, Simcha Therapeutics, among others often when they are no more than an idea. We are fanatical about helping the rare inventor who is compelled to build their own enduring business. If you or someone you know has a great idea or company in life sciences, Axial would be excited to get to know you and possibly invest in your vision and company . We are excited to be in business with you — email us at info@axialvc.com
Julia Schnabel's research is around medical image computing and machine learning. Focusing on intelligent imaging solutions and computer aided evaluation, including complex motion modeling, image reconstruction, image quality control, image segmentation and classification, applied to multi-modal, quantitative and dynamic imaging.
- AI-enabled imaging has many challenges, such as poor quality data, artifacts, patient motion, lack of large annotated datasets, legal and logistical frameworks, differences across different scanning systems, domain shift, and lack of interpretability.
- Reconstruction and segmentation are two important steps in medical imaging. Recently, a joint motion reconstruction segmentation method was developed that can improve the signal to noise ratio and other aspects of the image quality.
- To classify good and bad quality images, networks can be used in a supervised fashion, similar to classifying cats and dogs in an image.
- To bridge the gap between high quality images from UK Biobank and images from clinical settings, where motion artifacts are more likely to occur, we need to take into account the poorer quality images and make them look high quality.
- Image quality control is an important part of medical imaging, as it allows us to distinguish between good and bad images, and to restore images for downstream tasks such as segmentation and classification.
- When building complex architectures, empirical testing and trials should be conducted to compare the methods against clinicians’ work, and benchmarking should be done to measure performance. Regulatory approval is also necessary to ensure accuracy and safety.
- Julia suggested that it might be possible to learn the optimal loss function for a problem, rather than just using a mean squared error term or instructional similarity index. She also discussed the importance of having a modular approach, so that each individual component can be tested separately and justified to the clinician. However, she noted that there is still a risk of introducing errors or losing information.