Computer Vision News - March 2020

2 Summary AI in Medical Imaging 30 Who is l iable when a machine learning technique makes a decision? How should decisions be reviewed? Should machine learning check i f the doctor has missed something, or vice versa? The answers to these questions are not total ly clear yet, but progress continues unabated. Al though we have been talking more general ly about medical computer vi s ion, Orcun’s group al so has a focus on ul trasound imaging and appl icat ions, including developing novel imaging contrasts. One of thei r main l ines of research i s t issue biomechanical character izat ion . “Ti ssue biomechanics change wi th pathology,” Orcun explains. “Bas ical ly, pathology often makes i t st i f fer. Since ancient Greek and Egypt ian t imes , palpat ion has been used to feel t i ssue elast ici ty. Even today, i f you go to the doctor, they wi l l f i rst push on your tummy and i f something i s st i f f, that ’s not good. For breast and prostate cancer, the f i rst thing they wi l l do i s palpate and check i f something i s hard.” In hi s group, they use ul t rasound sequenc ing techniques , such as ul t rasound t ime of f l i ght data and image reconst ruct ion techniques , to do what i s ca l led an ul t rasound computed tomography. Thi s helps for mapping loca l speed and contribution. The technology needed is accessible to al l , so it ’s not a niche market anymore, and the computer vision community is very open in terms of code sharing. Most of the developed techniques are avai lable publ icly onl ine. “Now, many of these things are avai lable,” Orcun tel ls us . “I would say mostly in medical image computing – the MIC area rather than CAI – but sti l l there is a lot of data avai lable. That means many people can contribute. Then there are these large chal lenges that mean you can easi ly show that method X is better than al l the other methods, which makes it easier to bring i t into hospitals.” But there are sti l l unanswered questions. Once these methods are implemented in the real world, who takes responsibi l ity for them? Musculoskeletal modeling and functional biomechanical simulation of the shoulder by Fabien Pean & Carlos Oliveira