Vascular graph network for ovarian lesion classification using optical-resolution photoacoustic microscopy

TL;DR

This paper introduces a vascular graph network to classify ovarian lesions using microscopic imaging, improving accuracy in distinguishing cancer from benign conditions.

Contribution

The novel vascular graph network (VGN) leverages vascular structure for lesion classification with high accuracy using minimal tissue sampling.

Findings

VGN achieved 79.5% accuracy in diagnosing ovarian cancer.

The model provided stable predictions from small sampling areas (3 mm× 0.12 mm).

Five-class classification accuracy reached 73.4% using vascular graph data.

Abstract

Diagnosing ovarian lesions is challenging because of their heterogeneous clinical presentations. Some benign ovarian conditions, such as endometriosis, can have features that mimic cancer. We use optical-resolution photoacoustic microscopy (OR-PAM) to study the differences in ovarian vasculature between cancer and various benign conditions. In this study, we converted OR-PAM vascular data into vascular graphs augmented with physical vascular properties. From 94 ovarian specimens, a custom vascular graph network (VGN) was developed to classify each graph as either normal ovary, one of three benign pathologies, or cancer. We demonstrated for the first time that, by leveraging the intrinsic similarity between vascular networks and graph constructs, VGN provides stable predictions from sampling surface areas as small as 3 mm× 0.12 mm. In diagnosing cancer, VGN achieved 79.5 % accuracy and an area under the receiver operating characteristic curve (AUC) of 0.877. Overall, VGN achieved a five-class classification accuracy of 73.4 %.

Figures