Research
Introduction
Current computer vision algorithms are quite limited in detecting anomalies outside of preconfigured constraints. Instead of starting w/ pattern classification from scratch, we want to classify and tag the data manually. This allows us to take it a step further and build data models regarding the sentiment for anomaly detection, and preferred classification pipelines.
Project Aim
Our aim is to use these classifications to build further data models and research based on the tagging and survey results. The related research can then be open and published on preprints or journals referencing this study. Eventually through further phases, and perhaps even more Zooniverse projects, we may make more and more complex data models and algorithms as a result.
The Process
The process is simple. Once the survey is done, we input the data to an AI. We may also input other labeled data from open repos depending on the survey result's data quality. Once we input the data, we create a pipeline in which the AI learns about image regions, sorting, and tagging images. Depending on the results we get, we start making data models and may even move into a secondary survey phase if needed.
Project Purpose
Having an AI that can take care of simple repetitive tasks can help radiologist better allocate their time towards doing more complex diagnosis and procedures. Outside of this, we can create research models and algorithms for a variety of areas outside of tagging, including even image quality improvement. Lastly, a large portion of the world doesn't have access to medical imaging. Technologies such as these help make radiology more accessible while saving more lives in the long run. This is true whether working with radiologist to reduce human error, giving radiologist more time for costly tasks, or utilizing computer vision to improve image quality.
Dataset
All data is sourced from the following Kaggle Dataset.
For usability, special thanks to Paul Mooney on Kaggle for making this public data set.
Sources of Inspiration
- Aidoc
- DetAll
- Enlitic
- Qure.ai
- Awesome-DeepRadiology
- X-Ray Images Enhancement
- Enhance Local Contrast (CLAHE)
The above is a list of some educational sources or projects that may inspire you further in this subject matter.
Regarding Publishing
Any research that our research team publishes that directly references this data, should be expected to be released in the "Research Tab" (When Available). This includes preprints, journal entries or data repositories on places like Figshare, Sourceforge, and GitHub. This includes research whether it is in the form of a software tool or a literary review describing the results.