Up till 02/22/2021, VinBigdata Chest X-ray Abnormalities Detection Challenge attracts 695 talented teams worldwide. (The rankings are constantly updated in real time on Kaggle – The world’s largest online platform for Artificial Intelligence and Data Science community.)
The top three positions is belonging to international teams with members from Argentina, India, US, and Russia. However, this ranking is calculated based on 10% of the test data, and the final one will be based on the other 90%, so the final standings may be different. Therefore, there are still many chances of winning the first prize of 20,000 USD and the prize of 8,000 USD for the best Vietnamese team.
The challenge is practical, requiring contestants to automatically localize and classify 14 types of thoracic abnormalities from chest radiographs. If successful, you’ll help build what could be a valuable second opinion for radiologists. An automated system that could accurately identify and localize findings on chest radiographs would relieve the stress of busy doctors while also providing patients with a more accurate diagnosis.
In the nearly two months since the launch, hundreds of questions have been discussed and received enthusiastic contributions from the international community. The content of discussion revolves around the multi-format data processing, problem solving methods integrating identification and classification of abnormal points, explaining the evaluation method mAP@0.4, etc.
“A dataset consisting of 18,000 X-ray images that VinBigdata shares through the competition is one of the most detailed X-ray datasets ever. Moreover, lesion areas have been annotated by 17 experienced radiologists instead of simple classification label method. ” AI engineer Nguyen Ba Dzung, one of the Kaggle Masters from VinBigdata’s Medical Imaging Department, said.
With more than 3 years of experience taking part in competitions on the Kaggle platform, reaching the top of no less than 10 contests and excelling with a solo performance that won the Top 1 worth $ 8,000 at the Global Wheat Detection 2020, Mr. Dzung shares 4 steps to solve image processing problems as below:
- Step 1: Data processing. Based on the training data set provided by the organizers (for example, 15,000 lung X-rays with labeled results), the team will divide it into smaller data sets to build and evaluate models. Usually, the split ratio is 80% (training) and 20% (validation). Then, transform image such as rotate, flip, and change brightness to increase the diversity of the data.
- Step 2: Model training: It is necessary to choose the most suitable model for the object detection problem in the image. For example, the most commonly used models today are YOLO (You Only Look Once), EfficientDet, FasterRCNN. After selecting the appropriate models, the team will run numerous different models simultaneously to conduct the training.
- Step 3: Combining model: With the same question, consulting many different opinions will give the most accurate and comprehensive answer. Likewise, when solving the Object Detection problem, in order to get information of different models, it is necessary to perform coordinate transformations of the bounding box (average coordinate, delete the intersecting bounding box …). For teams, each team member can process the models independently, then use the cluster algorithm to combine the models. Therefore, the sooner a team participates in the competition, the more members it has, the more advantages in terms of the number of models as well as the time it takes to build, combine and test the model.
- Step 4: Denoising. To increase the accuracy of the model, the last but equally important step is to remove the noise data from the training set. The teams should combine several methods of filtering and processing to make the data cleaner.
VinBigdata Chest X-ray Abnormalities Detection Challenge was organized on the Kaggle platform with a total prize of 50,000 USD. The contest provides Vietnamese datasets, including 15,000 chest X-ray images for training and 3,000 images for evaluation. The best team will win a prize of 20,000 USD. In addition, 8,000 USD will be awarded to the Vietnamese team with impressive performance.
On March 31, 2021, the competition will officially close. Only more than 1 month for teams to compete in the Chest X-ray Detection contest. Do not miss the opportunity to take part in an international competition.