Attentive CT Lesion Detection Using Deep Pyramid Inference with Multi-Scale Booster

This repo is a implementation of "Attentive CT Lesion Detection Using Deep Pyramid Inference with Multi-Scale Booster"

Our code based on open-mmlab's mmdetection 18593f6. We made some changes on mmdetection.

Introduction

In this paper we propose a Multi-Scale Booster (MSB) with channel and spatial attentions integrated into the backbone Feature Pyramid Network(FPN). In each pyramid level, the proposed MSB captures fine-grained scale variations by using Hierarchically Dilated Convolutions (HDC).Meanwhile, the proposed channel and spatial attention modules increase the network attention on the feature responses to facilitate the lesion detection process.

Main Result

method	backbone	number of slices	FPs per image
			0.5	1	2	4	8
3DCE[1]	VGG-16	3	0.569	0.673	0.756	0.816	0.858
	VGG-16	9	0.593	0.707	0.791	0.843	0.878
	VGG-16	27	0.625	0.737	0.807	0.857	0.891
Faster Rcnn	Resnet-50	3	0.56	0.677	0.763	0.832	0.867
FPN	Resnet-50	3	0.621	0.728	0.807	0.864	0.89
FPN + MSB	Resnet-50	3	0.67	0.768	0.837	0.89	0.91

Install

Please refer to INSTALL.md for installing mmdetection.

(It is recommended that you install mmcv and mmdetection use pip install . -e --user, then you can modify the code)

Prepare DeepLesion dataset

Download deeplsion dataset[2] in https://nihcc.app.box.com/v/DeepLesion

Usage

Train

python train.py --cfg <config_path>

Test

Run test.ipynb(in jupyter) for test.

References

1. K. Yan, M. Bagheri, and R. M. Summers, “3D Context Enhanced Region-based Convolutional Neural Network for End-to-End Lesion Detection,” in MICCAI, 2018 (arXiv)

2. K. Yan, X. Wang, L. Lu, and R. M. Summers, “DeepLesion: Automated Mining of Large-Scale Lesion Annotations and Universal Lesion Detection with Deep Learning,” J. Med. Imaging, 2018. (paper)