Content area
Full Text
Abstract:
Adnexal masses are common, yet challenging, in gynecological practice. Making the differential diagnosis between their benign and malignant condition is essential for optimal surgical management, but reliable pre-surgical differentiation is sometimes difficult using clinical features, ultrasound examination, or tumor markers alone. A possible way to improve the diagnosis is using artificial intelligence (AI) or logistic models developed based on compiling and processing clinical, ultrasound, and tumor marker data together. Ample research has already been conducted in this regard that medical practitioners could benefit from. In this systematic review, we present logistic models and methods using AI, chosen based on their demonstrated high performance in clinical practice. Although some external validation of these models has been performed, further prospective studies are needed in order to select the best model or to create a new, more efficient, one for the pre-surgical evaluation of ovarian masses.
Keywords: artificial intelligence; deep learning; diagnosis; image interpretation; computer-assisted
Introduction
Ovarian tumors commonly present diagnostic challenges in gynecological practice. For instance, it is important to make the difference between a benign and a malignant ovarian tumor before surgery, especially in young patients who desire fertility. While benign masses can be treated conservatively [1] or by surgical removal thorough minimally invasive surgery [2], masses suspected of being malignant should be referred to a tertiary care center which may already be dealing with high numbers of ovarian cancer cases [3]. Therefore, accurate diagnosis is essential for planning appropriate patient management [4].
The diagnostic performance is in direct correlation with the experience of the physician [5]. The first imagistic step in evaluating an adnexal mass is ultrasonography (US), but US reports are sometimes misleading and confusing for the clinician [6]. To distinguish between benign and malignant tumors by means of US examination is not always easy because, for example, the imagistic features of borderline tumors overlap significantly with those of invasive epithelial cancers. In order to overcome such a difficulty, simple rules [7] or scoring systems were introduced with reportedly good results [8]. However, their application is not superior to the subjective impression of an experienced examiner [9], so further alternatives to improve imagistic diagnosis are still required.
One possible way to improve the accuracy of US diagnoses of ovarian tumors is to...