Utilizing Language Technology for Radiology Assessment
240,000 breast cancer diagnoses are made yearly, and about 3% of women die from breast cancer yearly. The sooner it is caught, the more effective the treatment and survival odds.
Once an abnormality has been detected, ultrasounds are used to further investigate. This is a crucial part of the process, determining whether or not tumors are malignant. Our project focuses on the screening, detection, and identification of malignant tumors within breast tissue.
61% of radiologists, the doctors who diagnose ultrasounds, report burnout. Many cite an exorbitant amount of bureaucratic work and charting as a leading cause. With breast cancer screening a crucial part of preventative health care, ultrasound screenings can contribute to this burnout.
ULTRAsound can alleviate burnout by generating tumor diagnosis reports from ultrasound images, helping radiologists focus on caring for those who need them most.
We first analyze breast tissue to isolate for any masses. If any are present, we further diagnose the mass as benign or malignant through CNN and CV techniques. The results of this diagnosis are then formatted into a radiology report through NLP techniques via open-source LLMs trained in medicine.