Medical Image Annotation: An Overview

September 10, 2023

Machine Learning is disrupting the world of medicine and healthcare, allowing professionals to diagnose patients better and faster than before. However, any Medical AI ML model training needs high-quality annotated medical images in large quantities. Here is where medical data labeling becomes essential. 

This blog explores everything you need to know about medical data annotation. If you are a pro already and have a medical image annotation project at hand, sign up for iMerit’s Ango Hub and start labeling your medical images.

What is Medical Image Annotation?

Medical data labeling is the process of annotating medical data, be it imaging data such as CT scans, X-rays, MRIs, ultrasounds, and retina fundus shots, so that healthcare professionals and machine learning algorithms can accurately interpret and diagnose medical conditions, track disease progression, and make informed treatment decisions.

The healthcare industry also requires other types of data labeled, such as document data like medical records in PDF or PNG/JPG formats. Medical data labeling can include audio labeling, such as patient conversations or cough sounds. This blog will focus on medical imaging.

AI teams use labeled data to train their ML models, which, once trained, can then automatically detect objects, lesions, tumors, and other abnormalities.

Getting Medical Images Ready for Labeling

In order to train a machine learning model that can produce dependable results, it is crucial to provide it with a substantial amount of high-quality labeled data. Frequently, obtaining this data, even in its unlabeled form, can be challenging. Additionally, when you have access to the data, there are a couple of considerations to keep in mind.

Variety of Datasets

Ensuring data diversity is crucial; it should not originate solely from one source or exhibit uniformity in appearance. The goal is to make the model as robust as possible to handle a wide range of real-world scenarios. If the model was trained on a subset of data closely resembling each other, it might struggle when presented with diverse data.

In essence, incorporate data from various sources, stages, institutions, or locations to enhance the model’s adaptability to different situations.

The Dataset Vetting Process

We recommend splitting your dataset into training, validation, and testing, where the training dataset should comprise about 80% of your data.

First, train your model with the training set and then evaluate the results on a small validation set. Look at the results that come out of the validation set. Are they to your satisfaction? 

Likely, they will need some tweaking. Tweak, then train again, and validate again. Repeat until you are satisfied with the validation results.

Once you are happy with the validation results, test your results against the test dataset. It will be your final model benchmark.

Size of your Dataset

Recent developments in ML have shown that quality is as important as quantity in training models. It means that a smaller but high-quality set will usually perform equally or even better than a large set of lower quality. That said, if you have the option to enlarge your dataset, we highly recommend doing so, as model results will improve significantly.

Format of your Dataset

The two most common medical imaging formats around are DICOM and TIFF. DICOM, especially, is the industry standard for radiologists. DICOM and TIFF files can optionally contain multiple images, slices, and metadata regarding the patient and the image itself. Good medical image annotation platforms will support both these formats, and the iMerit Radiology Editor, powered by the Ango Hub, can automatically remove identifying information from both metadata and the image itself on upload.

What makes medical image annotation different from others?

Labeling images for healthcare is an altogether different endeavor compared to regular image annotation. Here are some things that are different:

Data Availability

While regular images are often freely available or behind a standard NDA, medical imaging is usually protected by strict data processing agreements. It is mainly to protect the privacy of the patient. Obtaining medical imaging data is usually a longer process than other data types.

Technical Differences

Regular images only have one layer, are of small size, and have a low bit depth. Medical images often have multiple layers (slices), are huge, and have a higher bit depth.

Further, the labeler profiles for both will be different, where the annotation of medical images demands expertise from specialized healthcare professionals. These experts are used to certain UI and UX paradigms. Therefore, when choosing a data labeling platform, it is critical to note whether medical professionals can easily use its keyboard controls and UI.

Picking the Medical Image Annotation Tool for You

DICOM viewers with annotation capabilities abound in the market. One notable open-source option, for example, is 3D Slicer. DICOM viewing tools, however, are not optimized for ML model training. Sometimes, it is impossible to use the labels from these viewers in machine learning due to a lack of instance IDs and structured export formats. You must use a professional medical imaging labeling tool to train and develop a neural network.

Answer below for the image annotation solution you use or are choosing:

  • Does the solution support medical formats such as DICOM and TIFF?
  • Does it support the labeling features you are looking for?
  • Is the UX easy to use and suitable for medical use?
  • Is the export format easy to use in ML model training?
  • Does the tool provider have a medical data labeling service to enhance your workforce?

At iMerit, we are proud to offer all of these and more. If you need data annotation consulting and support, contact us today.