One of my most impactful projects involved annotating medical images for a healthcare startup developing AI tools to detect early signs of lung cancer. My role was to label CT scans by identifying nodules, differentiating between benign and malignant growths, and ensuring consistency across thousands of images.
We used tools like Labelbox to manage the data and ITK-SNAP to view scans in 3D. I worked closely with radiologists who gave us clear rules. When the medical guidelines changed halfway through, I had to go back and update hundreds of scans. Labelbox made this easier by letting me filter and re-check only the affected cases.
To keep quality high, we used a two-step review (peer check + radiologist check). Any unclear cases were tagged for group discussion in weekly meetings. We met deadlines by splitting scans into batches, tracking progress on dashboards, and setting daily goals.