Gravity Network for end-to-end small lesion detection

TL;DR

This paper introduces a novel one-stage end-to-end detector specifically designed to detect small lesions in medical images using a new type of pixel-based anchor that dynamically moves towards the targeted lesion for detection.

Abstract

This paper introduces a novel one-stage end-to-end detector specifically designed to detect small lesions in medical images. Precise localization of small lesions presents challenges due to their appearance and the diverse contextual backgrounds in which they are found. To address this, our approach introduces a new type of pixel-based anchor that dynamically moves towards the targeted lesion for detection. We refer to this new architecture as GravityNet, and the novel anchors as gravity points since they appear to be "attracted" by the lesions. We conducted experiments on two well-established medical problems involving small lesions to evaluate the performance of the proposed approach: microcalcifications detection in digital mammograms and microaneurysms detection in digital fundus images. Our method demonstrates promising results in effectively detecting small lesions in these medical imaging tasks.

Figures (5)

• Figure 1: Gravity-points distribution: on the left, the feature grid of size $K \times K$; in the middle, the entire image $H \times W$; on the right, the feature map $H_{FM} \times W_{FM}$.
• Figure 2: GravityNet architecture is composed of a backbone (blue) and two subnetworks, attached to the backbone output, one for classification task (orange) and one for regression task (green). The output is a representation of the gravity points in the grid pattern at training time and the subsequent attraction behavior towards the lesion at inference time. Gravity points in light blue correspond to positive candidates trained to collapse toward the ground truth in light green
• Figure 3: Hooking process where gravity points (light blue) are hooked to a lesion (light green)
• Figure 4: An example of NMS: on the left, gravity points and corresponding boxes (light blue) hooked to a lesion (green); on the right, the final candidate corresponding to the gravity point with the highest score (blue)
• Figure 5: Examples of initial gravity-points configurations represented in a reference window $K \times K$, where: (a)step = 5 (b)step = 6 (c)step = 10 (d)step = 15 (e)step = 30