I have a sizeable, fully-anonymised collection of CT scans and I’m ready to turn them into a working machine-learning solution. The exact clinical task—whether automated diagnosis, organ or lesion segmentation, or image quality enhancement—can be finalised together once you have reviewed the dataset and its labels, but the model must be trained exclusively on CT data. All work will be done in Python, and I’m happy to use PyTorch, TensorFlow, or another modern deep-learning framework if you feel it suits the problem best. I’ll provide the raw DICOM files (or NIfTI, if you prefer) along with any existing annotations. Your job is to design the training pipeline, build and tune the model, and package a reproducible inference script that can run on fresh scans without hassle. Deliverables I need at hand-off: • Clean, well-commented source code and environment file • Trained model weights (or a checkpoint) • Inference script that accepts untouched CT volumes and outputs the chosen predictions • Short report explaining architecture choices, performance metrics, and instructions for further fine-tuning Accuracy, stability and clear documentation are the acceptance criteria. If this sounds like a fit, let’s discuss the dataset, the clinical objective and any pre-processing you’d recommend so we can get started right away.