Task


The purpose of this challenge is to compare approaches of ophthalmic disease classification in color fundus images. Participant will have to submit classification results of eight categories for all the testing data. For every category, a classification probability (value from 0.0 to 1.0) denotes risk of a patient diagnosed with corresponding category.  


Result Submission
Submissions of results will be accepted in CSV file (named as [{Team Name}_{ODIR}.csv] example filename: “XYZ_ODIR.csv”). It must have a header (pid denotes patient id, must be same with provided patient id in ODIR-5K database) and should look like the following:

pid,N,D,A,G,C,H,M,O
1,0,0.8,0,0,0,0,0,0.9
2,1.0,0,0,0,0,0,0,0

Performance Evaluation
Submissions are scored based on three metrics: mean per-class accuracy (mPCA),  F-1 socre and AUC value. mPCA firstly computes accuracy for each class, then takes the mean of all the eight accuracy values. The threshold is 0.5.

A final score is the average of the above metrics.




该竞赛的目的是比较基于彩色眼底图像进行眼科疾病分类的不同方法。 参与者必须提交所有测试数据集的八个类别的分类结果。 对于每个类别,分类概率(值从0.0到1.0)表示患者被诊断为具有相应类别的可能性/风险。


结果提交

结果提交只接受CSV文件格式(名为[{Team Name} _ {ODIR} .csv], 示例文件名:“XYZ_ODIR.csv”)。 它必须有一个标题栏(其中pid表示患者的ID,和数据库的患者ID一一对应),形式如下:

PID,N,D,A,G,C,H,M,O
1,0,0.8,0,0,0,0,0,0.9
2,1.0,0,0,0,0,0,0,0


性能评估

对提交结果用三个指标进行评分:平均类别准确率(mPCA)F-1分数AUC面积。mPCA首先计算每个类的准确,然后取所有八个准确率的平均值。阈值是0.5。

最终得分是上述指标的平均值。