The project on “Deep learning for improved detection of premalignant lesions in the Fallopian tube ” aims to improve the diagnostics of precursor lesions to high grade serous carcinoma (HGSC), the most common and lethal form of ovarian cancer. The most relevant precursor lesion is known as Serous Tubal Intraepithelial Carcinoma (STIC). Consistent and reproducible diagnostics of these STIC lesions is important for three reasons. Firstly it is important for individual patient care. We know that an isolated STIC holds a significantly increased risk of peritoneal carcinomatosis. Therefore, diagnosing STIC holds important prognostic information for individual patients and may in the future also have treatment implications. Secondly reliable STIC diagnosis is important for better understanding the oncogenesis of high grade serous carcinoma. The hypothesis of STIC as the precursor to HGSC dates from 2001. Though we have learned a lot about HGSC since then, the etiology is still not fully unraveled. Thirdly, reliable STIC diagnosis plays and important role in investigating new risk reducing strategies for women at an increased risk of ovarian cancer, as reliable diagnostics are a prerequisite in safely offering these new strategies.

Diagnosing STIC however, is a challenging task for pathologists. It is a relatively rare diagnosis, which the pathologist may only encounter infrequently. The lesions are often very small and there is a high interobserver variability, as was shown in two previous reproducibility studies. We therefore aim to improve STIC diagnostics by developing and validating a deep learning algorithm, that can fully automatically analyze all available tissue sections, and aid the pathologist in identification of premalignant Fallopian tube epithelium. In doing so we aim to achieve a more accurate and consistent recognition of STIC.

Funding

KWF