IAFI-FCOS: Intra- and across-layer feature interaction FCOS model for lesion detection of CT images

TL;DR

This work tackles CT lesion detection by addressing loss of detail during multi-scale feature fusion and limited context modeling around small lesions. It introduces IAFI-FCOS, featuring an ICAF-FPN neck that integrates intra-layer context augmentation via dilated attention and across-layer feature weighting via dual-axis attention, followed by adaptive feature fusion. The approach yields state-of-the-art performance on a private pancreatic CT dataset and strong generalization on the public DeepLesion dataset, improving both detection accuracy and sensitivity for small lesions. By enhancing cross-scale interactions and context-aware representations, the method promises faster and more reliable automated screening to support radiologists in early cancer detection.

Abstract

Effective lesion detection in medical image is not only rely on the features of lesion region,but also deeply relative to the surrounding information.However,most current methods have not fully utilize it.What is more,multi-scale feature fusion mechanism of most traditional detectors are unable to transmit detail information without loss,which makes it hard to detect small and boundary ambiguous lesion in early stage disease.To address the above issues,we propose a novel intra- and across-layer feature interaction FCOS model (IAFI-FCOS) with a multi-scale feature fusion mechanism ICAF-FPN,which is a network structure with intra-layer context augmentation (ICA) block and across-layer feature weighting (AFW) block.Therefore,the traditional FCOS detector is optimized by enriching the feature representation from two perspectives.Specifically,the ICA block utilizes dilated attention to augment the context information in order to capture long-range dependencies between the lesion region and the surrounding.The AFW block utilizes dual-axis attention mechanism and weighting operation to obtain the efficient across-layer interaction features,enhancing the representation of detailed features.Our approach has been extensively experimented on both the private pancreatic lesion dataset and the public DeepLesion dataset,our model achieves SOTA results on the pancreatic lesion dataset.

Figures (6)

• Figure 1: CT scan visualization of the pancreas dataset. (a) easy to detect: tumor features are distinct and have clear boundaries. (b) hard to detect: tumor boundaries are fuzzy and the target is small, a situation that is often difficult for the network to identify.
• Figure 2: Overview of the network architecture of the IAFI-FCOS detection framework, which mainly consists of three components: (a)a backbone network for feature extraction, (b)the ICAF-FPN neck and (c)the object detection head network. The ${C_{i}}$, ${P_{i}}$, ${IF_{i}}$, ${AF_{i}}$ and ${L_{i}}$ represent feature maps, the ${W_{i}}$ indicate learnable weights.
• Figure 3: The structure of ICA block.
• Figure 4: The structure of AFW block. The left and right sides of the figure represent the specific processes of AFG and Split FW, respectively.
• Figure 5: The FROC curves of various methods for detection on the Deeplesion dataset.