Skip to main content

Table 2 Mean segmentation Dice score for the different targets and evaluations in both SCAPIS and IGT Datasets

From: Automatic segmentation of large-scale CT image datasets for detailed body composition analysis

Segmentation target

Experimental results

SCAPIS

IGT

ResUNET

UNET++

Ghost-UNET

Ghost-UNET++

ResUNET

UNET++

Ghost-UNET

Ghost-UNET++

CV

Test

CV

Test

CV

Test

CV

Test

CV

Test

CV

Test

CV

Test

CV

Test

Liver slice

Liver accurate

0.989

0.994

0.986

0.989

Liver crude

0.917

0.928

0.986

0.985

0.969

0.971

0.979

0.978

0.961

0.967

0.987

0.985

0.982

0.979

0.982

0.981

Spleen

0.966

0.993

0.976

0.978

Abdomen slice

VAT

0.967

0.968

0.973

0.974

0.967

0.961

0.972

0.978

0.967

0.971

0.973

0.976

0.965

0.967

0.973

0.970

IPAT

0.955

0.954

0.979

0.973

0.943

0.935

0.957

0.951

RPAT

0.966

0.951

0.975

0.966

0.952

0.938

0.954

0.942

SAT

0.993

0.993

0.990

0.996

0.991

0.992

0.994

0.994

0.992

0.993

0.995

0.996

0.993

0.992

0.995

0.994

DSAT

0.946

0.941

0.972

0.973

0.926

0.944

0.951

0.951

SSAT

0.955

0.947

0.968

0.959

0.913

0.921

0.925

0.928

Spine bone marrow

0.979

0.949

0.993

0.984

0.979

0.976

0.985

0.979

Skeleton muscle

0.971

0.976

0.988

0.982

0.973

0.972

0.974

0.975

Thigh slice

IMAT

0.909

0.918

0.927

0.931

0.904

0.906

0.910

0.912

0.906

0.913

0.914

0.916

0.897

0.895

0.905

0.904

Muscle

0.996

0.996

0.996

0.995

0.991

0.993

0.994

0.996

0.996

0.995

0.996

0.996

0.994

0.993

0.995

0.993

SAT

0.988

0.988

0.992

0.991

0.985

0.986

0.987

0.988

0.989

0.989

0.991

0.992

0.985

0.983

0.988

0.987

Average dice score

0.964 ± 0.025

0.959 ± 0.024

0.981 ± 0.017

0.976 ± 0.017

0.961 ± 0.028

0.958 ± 0.027

0.968 ± 0.025

0.964 ± 0.026

0.968 ± 0.031

0.972 ± 0.028

0.976 ± 0.028

0.977 ± 0.028

0.969 ± 0.034

0.968 ± 0.034

0.973 ± 0.032

0.972 ± 0.032

  1. Mean dice score of the respective targets and experiments is presented. Average dice scores are in (mean ± standard deviation) and boldface represents the best scores. The models were trained and validated individually for each reference segmentation target
  2. CV cross validation