ViCOROB presented Artificial Intelligence in Breast Cancer Diagnosis
Profound improvements have been introduced by artificial intelligence techniques in general, and deep learning in specific, to the medical image processing and analysis field in the last decade. That was due to the surge of data availability derived from internet spread and social media. One aspect of AI is referred to as Artificial Neural Networks (ANN), which in turn has several kinds including Generative Adversarial Neural Networks (GAN). Based on that, the research done by Basel Alyafi is initially on the use of GAN and other generative models in realistic medical image synthesis and analysis. That can grow to measure uncertainty estimation, which emphasises the robustness of a system, and survival analysis, where the time-to-event is predicted. The data he deal with can range from full COVID-19 CT/X-ray lung images to mammographic lesion patches. It is hypothesised that by incorporating those methods, the performance of conventional methods can be sharply enhanced.
From the 24th to 27th of May 2020, the 15th International Workshop on Breast Imaging – IWBI2020 took place in the Ku Leuven University, Belgium. IWBI is designed as a platform to present the latest physical-technical developments, quality control and clinical experiences with novel breast imaging technologies, including not only digital mammography and tomosynthesis, but also breast CT, MR, ultrasound, optical and molecular imaging. Additional topics include multimodality imaging, image processing and visualization, and computer aided imaging. IWBI brings together researchers, clinicians and representatives from industry, who are jointly committed to developing technology and other image-based tools for the early detection and subsequent management of breast cancer. The workshops are designed to help advance the fields of breast cancer and medical imaging through sharing scientific discoveries, best clinical practices, and industrial innovations. ( From the website of European Society of Breast Imaging)
At this conference, the medical imaging laboratory presented the work of the PhD student Basel Alyafi: “Quality analysis of DCGAN-generated mammography lesions”. During this online workshop, we showed that Deep Convolution Generative Adversarial Networks are able to synthesise realistic and diverse patches of breast mass and microcalcification lesions. That was shown by the help of two methods: 1) the 2D visualization of real and synthetic lesions, 2) an observer study where two expert radiologists generously participated to judge the realism of a sample of the generated lesions. The collaborations were mainly with Clinica Girona and Parc Tauli Hospital Universitari in Barcelona. This work was supported by the project of ICEBERG (ref. RTI2018-096333-B-I00) led by Dr. Robert Marti. The authors gratefully acknowledge the donation of the Titan X GPU used for this project by NVIDIA Corporation.
The 15th International Breast Imaging Workshop could not be held in Leuven due to the continuous spread of the COVID-19 virus at the time. The workshop panel did a great job with that regard where the talks were collected from the authors and then streamed live on a YouTube channel and simultaneously via Skype-for-business meetings. Discussion and question-and-answer sessions were held after each set of presentations to share ideas among the authors themselves and with the audience as well.
No Comment