This paper presents an innovative deep learning approach for the detection of precancerous colorectal polyps, aiming to improve diagnostic accuracy and efficiency. Colorectal polyps are early-stage growths that can develop into colorectal cancer, making their early detection crucial for reducing cancer incidence and mortality. The study evaluates three leading object detection algorithms—YOLOv3, FCOS, and Faster R-CNN—using a comprehensive dataset of CT scans from patients with colorectal polyps. The goal is to identify the most effective model parameters for detecting polyps in CT images, thereby advancing machine-assisted detection methods. The research explores the theoretical foundations of neural networks, convolutional neural networks (CNNs), and object detection algorithms. It discusses the role of deep learning in medical imaging, particularly in automating polyp detection with reduced human error. The study also examines the performance of different feature extraction networks, including ResNet50, ResNet101, and ResNeSt50, to determine their impact on detection accuracy. The experimental design involves training and evaluating detection models using a dataset of CT scans, with metrics such as mean average precision (mAP), precision, recall, and intersection over union (IoU) used to assess performance. The study finds that two-stage detection models, such as Faster R-CNN, generally outperform single-stage models in terms of accuracy. ResNeSt50 is identified as the optimal feature extraction network for the Faster R-CNN model, significantly improving detection accuracy. The research concludes that deep learning technologies, particularly CNNs, offer significant potential for improving the early detection and prevention of colorectal cancer. By leveraging advanced deep learning techniques, the study contributes to more accurate and efficient diagnostic tools for colorectal polyp detection, ultimately aiding in the reduction of cancer incidence and mortality. The findings highlight the importance of continued research in deep learning for medical imaging applications.This paper presents an innovative deep learning approach for the detection of precancerous colorectal polyps, aiming to improve diagnostic accuracy and efficiency. Colorectal polyps are early-stage growths that can develop into colorectal cancer, making their early detection crucial for reducing cancer incidence and mortality. The study evaluates three leading object detection algorithms—YOLOv3, FCOS, and Faster R-CNN—using a comprehensive dataset of CT scans from patients with colorectal polyps. The goal is to identify the most effective model parameters for detecting polyps in CT images, thereby advancing machine-assisted detection methods. The research explores the theoretical foundations of neural networks, convolutional neural networks (CNNs), and object detection algorithms. It discusses the role of deep learning in medical imaging, particularly in automating polyp detection with reduced human error. The study also examines the performance of different feature extraction networks, including ResNet50, ResNet101, and ResNeSt50, to determine their impact on detection accuracy. The experimental design involves training and evaluating detection models using a dataset of CT scans, with metrics such as mean average precision (mAP), precision, recall, and intersection over union (IoU) used to assess performance. The study finds that two-stage detection models, such as Faster R-CNN, generally outperform single-stage models in terms of accuracy. ResNeSt50 is identified as the optimal feature extraction network for the Faster R-CNN model, significantly improving detection accuracy. The research concludes that deep learning technologies, particularly CNNs, offer significant potential for improving the early detection and prevention of colorectal cancer. By leveraging advanced deep learning techniques, the study contributes to more accurate and efficient diagnostic tools for colorectal polyp detection, ultimately aiding in the reduction of cancer incidence and mortality. The findings highlight the importance of continued research in deep learning for medical imaging applications.