Skip to main content

Table 4 Experimental results of the NLP algorithm for each fracture type

From: Natural language processing of radiology reports for identification of skeletal site-specific fractures

Fractures

Sensitivity

Specificity

PPV

NPV

F1-score

Ankle

0.974

1.000

1.000

0.974

0.987

Clavicle

1.000

1.000

1.000

1.000

1.000

Distal Forearm

1.000

1.000

1.000

1.000

1.000

Face

0.760

1.000

1.000

0.806

0.864

Feet and Toes

0.960

1.000

1.000

0.962

0.980

Hand and Fingers

0.918

1.000

1.000

0.924

0.957

Other Spine Fractures

0.875

1.000

1.000

0.889

0.933

Patella

1.000

1.000

1.000

1.000

1.000

Pelvis

0.952

1.000

1.000

0.955

0.976

Proximal Femur

1.000

1.000

1.000

1.000

1.000

Proximal Humerus

1.000

1.000

1.000

1.000

1.000

Ribs

0.933

1.000

1.000

0.938

0.966

Scapula

1.000

1.000

1.000

1.000

1.000

Shaft and Distal Femur

0.800

1.000

1.000

0.833

0.889

Shaft and Distal Humerus

0.857

1.000

1.000

0.875

0.923

Shaft and Proximal Radius/Ulna

0.952

1.000

1.000

0.955

0.976

Skull

1.000

1.000

1.000

1.000

1.000

Sternum

1.000

1.000

1.000

1.000

1.000

Tibia and Fibula

0.944

1.000

1.000

0.947

0.971

Vertebral Body

0.675

1.000

1.000

0.755

0.806

Micro-Average

0.930

1.000

1.000

0.941

0.961