Surgical Blade Inspection

True story: A manufacturer produces thousands of ophthalmic surgical blades and these blades go through multiple visual inspections before packing but due to human error a wrong blade got packed under a different name and was shipped to the hospital.

A surgeon while operating on a patient gets the wrong blade.

This is a huge problem for the doctor, the patient and the manufacturer.

So I wanted to build a surgical blade inspection system using deep learning and computer vision to ensure no wrong blade gets shipped to the surgeon.

Surgical Blades:

Ophthalmic blades are broadly classified into 4 types based on their shapes:

1. Keratome

2. Implant

3. Sideport

4. Crescent

and has subcategories with varying sizes in the range of mm, so it is a very tricky problem to solve and that is where deeplearning based object detection model comes into play.

Setup:

1. We created a region where the blades can be inspected using a camera

2. Prototype: KV260 and Raspberry pi Camera:

The camera will be triggered by a PLC to capture an image for the AI model to classify.

Using this setup I collected data to train and test my AI model.

3. AI Software Prototype:

For this problem statement I used Yolov5m an object detection model, I trained and validated the model using the data I collected.

The Model achieved an impressive 95% accuracy on the test set and it can be further improved with more training images.

I created a simple Flask based UI to show how the software prototype works

Next Steps:

I hope to improve the model by collecting data from the factory setup.

And Deploy in their facility.