Master's Thesis: Artificial intelligence in the fight against breast cancer
Carlotta Ruppert has developed a deep learning model for the detection and classification of breast lesions. This is a valuable contribution to the early detection of breast cancer. The goal is to be able to use the technology directly during examinations in the future.
In Switzerland, around 6300 women are diagnosed with breast cancer every year. Early detection significantly improves treatment outcomes. Carcinomas are generally diagnosed with mammography and ultrasound, which is used especially for examining denser breast tissue. However, this method is associated with a higher false positive rate and requires very experienced radiologists. To support them in their work, Carlotta Ruppert has developed a deep learning model for the detection and classification of breast lesions for her Master's thesis.
"I am excited that my work will be used right there in the hospital and has great potential to improve breast cancer diagnosis in the long term."
Carlotta Ruppert
Working with real cases
The graduate conducted her thesis project in collaboration with b-rayZ, a spin-off of the University Hospital Zurich that aims to develop sustainable artificial intelligence (AI) solutions for breast cancer diagnosis that are accessible to all women. This collaboration gave the graduate access to over 3000 ultrasound images and patient records that were approved for use in research. "These are classified according to the Breast Imaging Reporting and Data System (BI-RADS), which is used worldwide. This system is used to classify lesions in breast tissue according to their risk of carcinoma", she explains. A doctor marked these structures on the ultrasound images and labelled them. Carlotta Ruppert then used this data to train her deep learning model. Since about 80 percent of the training and validation images only showed benign breast lesions, the model's loss function had to be weighted.
Nearly as good as the experts
Carlotta Ruppert then started a trial in which the model and two independent radiologists analysed a set of 154 ultrasound images. In this trial, the results achieved by the model had a similar level of accuracy as those of the human experts. "This means that the system will be a great help in diagnosing breast cancer", the graduate believes. To put this into practice, Carlotta Ruppert will start a doctorate at the University Hospital Zurich: "I am excited that my work will be used right there in the hospital and has great potential to improve breast cancer diagnosis in the long term."
Our project addresses a necessity in breast cancer diagnosis: to reduce variability in the interpretation of breast sonography. The new technology significantly advances existing solutions on the market by improving the accuracy of findings and through its integration into the radiologists’ workflow.
PD Dr Cristina Rossi, CEO/Co-Founder, b-rayZ AG