Detector

• 

  Hugging Face

• 

  Canva 1

• 

  Inspiration

• 

  GitHub

• 

  Acc Curves

Inspiration

Cancer is disease that claims about 10-15 millions lives per year, however numbers usually don't affect us till the time it isn't a close one who gets targeted by it. Last year, I lost my maternal grandfather, if his cyst near the stomach was detected earlier he would be still be today with us

What it does

My project takes in as input an image of a malignant and benign oncological disease and tells the possibility for skin cancer along with its type (Actinic Keratosis, Basal Cell Carcinoma, Dermatofibroma, Melanoma, Nevus, Pigmented Benign Keratosis, Seborrheic Keratosis, Squamous Cell Carcinoma, Vascular Lesion)

How we built it

I trained an Efficient Net Model on (kaggle's Skin Cancer ISIC dataset)[https://www.kaggle.com/datasets/nodoubttome/skin-cancer9-classesisic/data] by KatanSkiy using images from The International Skin Imaging Collaboration (ISIC), at first the accuracy wasn't good so I added a hint of data augmentation and a scheduler which slightly improved my models performance.

Challenges we ran into

Finding the right dataset for the model since too big would mean its trains slowly as I don't have access to GPUs while too small might be too biased or doesn't do anything at all
Hosting on Hugging Face
Learning Gradio
Which model works best out the many trained by me

Accomplishments that we're proud of

A training accuracy of about 90% and a testing accuracy of about 50% so hopefully people with possible skin cancer might get early detection and hence early treatment.
Model is deployed on hugging for all to leverage if needed

What we learned

To host on Hugging Face
How to tackle underfitting

What's next for Brain Tumor Detector

To tackle overfitting since training accu. is about 90% while the testing accuracy lags behind at 50%
Testing on bigger datasets
Changing or improving the base model

Built With

• computervision
• github
• gradio
• huggingface
• kaggle
• python
• pytorch