Surgical teams often face delays due to bottlenecks in the sterilization process, where staff handle instruments without assessing actual contamination levels. A deep learning-based method is introduced that detects microbial colonies on agar plates aiming to optimize sterilization workflows. Several state-of-the-art object detection models were trained and evaluated, including Faster R-CNN with ResNet-50 and MobileNet backbones, SSD300 with VGG16, FCOS with ResNet-50-FPN, and YOLOv11 using both nano and small variants. All models were tested on a dataset containing over 9,000 unannotated images captured at 10-minute intervals to provide a dense temporal view of microbial development. Annotations were generated using ColTapp, a colony analysis tool that outputs colony coordinates and radii, and any noise introduced by the automatic process was addressed through prior data preprocessing. Despite this challenge, YOLOv11 consistently delivered the best results in both detection accuracy and computational performance. Its growing adoption in real-time vision applications reinforces its effectiveness, mainly due to its substantial speed-accuracy tradeoff. The top-performing configuration achieved 99.1% precision, a recall of 91.7% and a F1 score of 95.3%. This solution presents a scalable and cost-effective alternative to manual contamination assessment. By integrating this approach into hospital sterilization workflows, healthcare systems could reduce instrument turnaround times, improve resource allocation, and strengthen infection control. Ultimately, this work paves the way toward smarter, AI-driven clinical operations that enhance efficiency and patient safety.