Skip to main content

Table 3 Correct detection rates for test data

From: Deep learning-based automatic-bone-destruction-evaluation system using contextual information from other joints

 

Intact

Non-intact

Total

Erosion

 PIP-IP

916/921 (99.5%)

99/109 (90.8%)

1015/1030 (98.5%)

 MCP

896/900 (99.6%)

108/130 (83.1%)

1004/1030 (97.5%)

 CMC-M

289/291 (99.3%)

113/121 (93.4%)

402/412 (97.6%)

 Wrist

393/393 (100.0%)

415/431 (96.3%)

806/824 (98.1%)

 All joints

2494/2505 (99.6%)

735/791 (92.9%)

3229/3296 (98.0%)

JSN

 PIP

508/513 (99.0%)

285/311 (91.6%)

793/824 (96.2%)

 MCP

782/796 (98.2%)

203/234 (86.8%)

985/1030 (95.6%)

 CMC

304/305 (99.7%)

309/313 (98.7%)

613/618 (99.2%)

 Wrist

197/197 (100.0%)

420/421 (99.8%)

617/618 (99.8%)

 All joints

1791/1811 (98.9%)

1217/1279 (95.2%)

3008/3090 (97.3%)

  1. Correct detection rates for intact (SHS = 0) joints, non-intact (SHS ≥ 1) joints, and the total for each target joint are shown. The numbers represent the correct detection/total cases (correct detection rates %). In erosion, “Wrist” represents the navicular, the lunate, the radius, and the ulna. In JSN, “Wrist” represents the multangular-navicular, the capitate-navicular-lunate, and the radiocarpal joint
  2. GT ground-truth, PIP proximal interphalangeal, IP interphalangeal, MCP metacarpophalangeal, CMC-M carpometacarpal joint of the thumb and multangular, JSN joint space narrowing, CMC carpometacarpal