I will build medical image segmentation and classification models with deep learning
AI and Machine Learning Engineer, Deep Learning, Computer Vision, NLP
About this Gig
Need a deep learning model for medical images? I build segmentation and classification models that outline lesions or classify scans.
What I build
Segmentation with UNet, 3D UNet, TransUNet and Swin UNet
Classification with EfficientNet and CNN ensembles
2D images and 3D volumes, such as retinal OCT and fundus photos
Grad CAM, SHAP and LIME explanations
What you get
Clean, commented PyTorch or TensorFlow code and trained weights
Dice, IoU, accuracy and ROC AUC on unseen patients or images
Overlay images showing correct, missed and extra regions
Multiple seeds and statistical tests in the top package
A plain English report
Proven results
79.23% mean Dice, retinal OCT fluid segmentation on 6 datasets
78.46% mean Dice, 3D model on full OCT volumes
91.49% accuracy, 0.9731 ROC AUC, diabetic retinopathy
Why choose me
Benchmarked against TransUNet and Swin UNet with their official code
Strong maths background (MS Mathematics) for sound statistics
Free feasibility check before you order
Updates at every step and help running the code after delivery
Send me a few sample images and your goal, and I will reply with a plan.
Expertise:
Image processing
•
Feature learning
•
Classification
Programming language:
Python
•
Colab
Tools:
Jupyter Notebook
•
OpenCV
•
TensorFlow
•
Colab
Frameworks:
Scikit-learn
•
Keras
•
PyTorch
•
Panda
Other Data Science & ML Services I Offer
FAQ
Which image types do you work with?
I have worked with retinal OCT scans and fundus eye photos in 2D and 3D. Send a sample of your images and I will confirm the approach.
Do my images need masks or labels?
Segmentation needs mask labels, classification needs a label per image. If you only have raw images, message me first.
How do you test the model?
I split by patient or image so test data is never seen in training, and report Dice, IoU or accuracy on that test set.
Can you compare with published models?
Yes. In the research package I compare against models such as TransUNet and Swin UNet using their official code.
Do I get the code and trained weights?
Yes, along with the overlay images and a results report.

