Skip to main content

Table 3 Sensitivities of the different methods in Retinopathy Online Challenge dataset at the various FP/image rates

From: Microaneurysm detection in fundus images using a two-step convolutional neural network

Free-response receiver operating characteristic results on Retinopathy Online Challenge dataset at average number of False positives per image

FPs/img

Sensitivity

Method

1/8

1/4

1/2

1

2

4

8

Classification method

 Proposed method

0.047

0.173

0.351

0.552

0.613

0.722

0.769

CNN

 Dashtbozorg [38]

0.435

0.443

0.454

0.476

0.481

0.495

0.506

RUSBoost

 Chudzik [29]

0.142

0.201

0.250

0.325

0.365

0.390

0.409

CNN

 Budak [28]

0.039

0.061

0.121

0.220

0.338

0.372

0.394

DCNN

 Javidi [40]

0.130

0.147

0.209

0.287

0.319

0.353

0.383

Discriminative dictionary learning

 Wu’s [39]

0.037

0.056

0.103

0.206

0.295

0.339

0.376

KNN

 Valladolid [42]*

0.190

0.216

0.254

0.300

0.364

0.411

0.519

GMM

 Waikato group [41]*

0.055

0.111

0.184

0.213

0.251

0.300

0.329

Bayesian

 Latim [25]*

0.166

0.230

0.318

0.385

0.434

0.534

0.598

Thresholding

 OkMedical [10]*

0.198

0.265

0.315

0.356

0.394

0.466

0.501

Dynamic thresholding

 Fujita Lab [43]*

0.181

0.224

0.259

0.289

0.347

0.402

0.466

ANN

  1. The quantity given in italic form in each FPs/Img column represents the best result
  2. * Indicate papers which use the full original dataset and others which use the cross-validation technique