IOTA-AI: AI-assisted automated detection of ovarian cancer on ultrasound imaging

Ovarian cancer is the seventh most common cancer in women and constitutes the most lethal gynecological malignancy. Timely diagnosis and appropriate referral to gynecological oncology expert centers is pivotal to improve patient outcomes. Ultrasonography (US) is a readily available, cheap and harmless technique, and is widely accepted as first-line imaging modality for assessment of ovarian masses. Currently, the ADNEX model is the best available ultrasound-based mathematical model to differentiate between benign and several types of malignant ovarian tumors. However, it relies on the ability of the US operator to reliably (manually) locate, delineate and measure the tumor area and its associated features. Previously, we developed automated feature detection methods in collaboration with ESAT-STADIUS and General Electric, where we demonstrated the effectiveness of a deep convolutional neural network (DCNN) approach. The objective of this C3 project is to further develop and validate a fully automated classification model for triaging patients with ovarian masses. Implementation of such a model in centers where specialized ultrasound operators are not available, would facilitate the early detection of ovarian cancer and positively impact patients’ survival.

Keywords:AI - Deep learning and Transfer Learning, Clinical decision support systems, Ultrasound Imaging, Ovarian Cancer

Disciplines:Cancer diagnosis, Machine learning and decision making