Tumor Segmentation with Swin Transformer and UNet++

I'm looking for an experienced AI/ML developer or researcher to implement a deep learning-based medical image segmentation model for the detection and segmentation of liver tumors from medical images (CT slices). The model should utilize a Swin Transformer encoder combined with a UNet++ decoder, aiming for high accuracy in identifying tumor regions in 2D image slices.



Skills Needed:

Proficiency in Python and deep learning frameworks, particularly PyTorch.

Experience with Transformer-based vision models (e.g., Swin Transformer via timm).

Knowledge of UNet, UNet++, or similar encoder-decoder architectures.

Strong background in medical image processing.

Familiarity with evaluation metrics like Dice coefficient, IoU, and binary cross-entropy loss.

Deliverables:

A fully working model combining Swin Transformer (encoder) with UNet++ (decoder) for tumor segmentation.

Complete, clean, and well-documented Python code.

Image preprocessing pipeline (e.g., resizing, normalization).

Training and validation code with visual output.

Evaluation script (Dice score, accuracy metrics).

Inference script to test on a single image and visualize the result.

A short technical report outlining:

Model architecture.

Training process and data handling.

Results and performance metrics.

Challenges faced and how they were handled.