[Paper Review] RCS-YOLO: A Fast and High-Accuracy Object Detector for Brain Tumor Detection

Ming Kang, Chee‐Ming Ting+2 more|arXiv (Cornell University)|Jul 31, 2023

Brain Tumor Detection and ClassificationNeuroscience29 references3 citations

TL;DR

This paper proposes RCS-YOLO, a novel YOLO-based object detector for brain tumor detection that integrates Reparameterized Convolution with Channel Shuffle (RCS) and One-Shot Aggregation (RCS-OSA) to enhance feature extraction and computational efficiency. On the Br35H dataset, RCS-YOLO achieves 114.8 FPS inference speed and 1% higher precision than YOLOv7, setting a new state-of-the-art performance in speed and accuracy.

ABSTRACT

With an excellent balance between speed and accuracy, cutting-edge YOLO frameworks have become one of the most efficient algorithms for object detection. However, the performance of using YOLO networks is scarcely investigated in brain tumor detection. We propose a novel YOLO architecture with Reparameterized Convolution based on channel Shuffle (RCS-YOLO). We present RCS and a One-Shot Aggregation of RCS (RCS-OSA), which link feature cascade and computation efficiency to extract richer information and reduce time consumption. Experimental results on the brain tumor dataset Br35H show that the proposed model surpasses YOLOv6, YOLOv7, and YOLOv8 in speed and accuracy. Notably, compared with YOLOv7, the precision of RCS-YOLO improves by 1%, and the inference speed by 60% at 114.8 images detected per second (FPS). Our proposed RCS-YOLO achieves state-of-the-art performance on the brain tumor detection task. The code is available at https://github.com/mkang315/RCS-YOLO.

Motivation & Objective

• To address the lack of high-performance YOLO-based models in brain tumor detection.
• To improve feature representation and computational efficiency in object detection for medical imaging.
• To develop a lightweight yet accurate detector suitable for real-time clinical applications.
• To achieve state-of-the-art performance on the Br35H brain tumor detection benchmark.

Proposed method

• The proposed RCS-YOLO integrates a Reparameterized Convolution with Channel Shuffle (RCS) module to enhance feature learning and reduce redundancy.
• A One-Shot Aggregation (RCS-OSA) module is introduced to efficiently fuse multi-scale features in a single forward pass.
• The architecture combines the strengths of YOLO’s real-time inference with improved feature cascade through channel-wise shuffling and reparameterization.
• The model is trained and evaluated on the Br35H dataset using standard object detection metrics including mAP and FPS.
• Reparameterization is applied during inference to merge batch normalization and convolutional layers, improving speed without accuracy loss.

Experimental results

Research questions

• RQ1 Can a reparameterized convolution-based YOLO architecture improve both speed and accuracy in brain tumor detection?
• RQ2 How does the integration of channel shuffle and one-shot feature aggregation affect feature representation in medical image detection?
• RQ3 Does RCS-YOLO outperform existing YOLO variants such as YOLOv6, YOLOv7, and YOLOv8 on the Br35H dataset?
• RQ4 To what extent does the proposed method reduce inference time while maintaining or improving detection accuracy?

Key findings

• RCS-YOLO achieves an inference speed of 114.8 FPS on the Br35H dataset, representing a 60% improvement over YOLOv7.
• The model improves precision by 1% compared to YOLOv7 while maintaining high mAP, demonstrating superior accuracy.
• RCS-YOLO outperforms YOLOv6, YOLOv7, and YOLOv8 in both speed and accuracy, achieving state-of-the-art performance.
• The integration of RCS and RCS-OSA modules enhances feature extraction and reduces computational overhead, contributing to faster inference.