I need a deep-learning model that can pinpoint and label regions of interest within medical images (primarily DICOM and high-resolution PNG). My end goal is to integrate the trained model into an existing Python pipeline for automated reporting, so accuracy and clear documentation are crucial. Scope of work • Curate or augment a balanced dataset if one is not already publicly available. • Design, train, and fine-tune an object-detection architecture (Faster R-CNN, YOLOv8, or a comparable state-of-the-art approach in PyTorch or TensorFlow). • Apply data-augmentation and transfer-learning strategies to counter class imbalance and limited sample sizes typical in medical imaging. • Validate performance with standard metrics (mAP@0.5, precision, recall) on a held-out test set and provide a concise model card summarising results. • Package the final weights, inference script, and environment file (requirements.txt or Conda YAML) so I can reproduce training and run batch inference on GPU. Acceptance criteria • Minimum mAP@0.5 of 0.85 on the test set. • Inference time ≤100 ms per 1024×1024 image on an NVIDIA RTX-class GPU. • Clean, commented code plus a README covering setup, training, and inference steps. If you have prior experience with medical object detection or have published models in this space, that will help speed things up. I am ready to start as soon as we agree on milestones and timeline.