Resultados TFG

Resultados DL3D

Resultados con Deep Learning con solo volúmenes 3D y con el modelo DenseNet121

Set	Batch Size	LR	Weight Decay	Dropout	Preprocesado	Accuracy	F1	TPR	TNR	G-Mean
TEST	4	0.0010	0.00000	0.3	(64,64,64) HU [-1000,400]	0.696	0.680	0.679	0.709	0.691
VALIDATION	4	0.0010	0.00000	0.3	(64,64,64) HU [-1000,400]	0.720	0.697	0.720	0.722	0.715
TEST	16	0.0010	0.00000	0.4	(64,64,64) HU [-1000,400]	0.680	0.687	0.747	0.619	0.674
VALIDATION	16	0.0010	0.00000	0.4	(64,64,64) HU [-1000,400]	0.650	0.640	0.676	0.627	0.650
TEST	16	0.0010	0.00000	0.3	(64,64,64) HU [-1000,400]	0.680	0.713	0.830	0.542	0.666
VALIDATION	16	0.0010	0.00000	0.3	(64,64,64) HU [-1000,400]	0.660	0.626	0.633	0.685	0.655
TEST	8	0.0010	0.00000	0.2	(64,64,64) HU [-1000,400]	0.656	0.664	0.733	0.592	0.652
VALIDATION	8	0.0010	0.00000	0.2	(64,64,64) HU [-1000,400]	0.650	0.620	0.631	0.664	0.643
TEST	16	0.0010	0.00001	0.4	(128,128,128) HU [-1350,150]	0.664	0.655	0.664	0.662	0.650
VALIDATION	16	0.0010	0.00001	0.4	(128,128,128) HU [-1350,150]	0.650	0.629	0.656	0.653	0.646
TEST	16	0.0010	0.00000	0.2	(64,64,64) HU [-1000,400]	0.656	0.648	0.697	0.620	0.645
VALIDATION	16	0.0010	0.00000	0.2	(64,64,64) HU [-1000,400]	0.660	0.624	0.609	0.702	0.650
TEST	32	0.0010	0.00000	0.2	(64,64,64) HU [-1350,150]	0.664	0.629	0.629	0.692	0.644
VALIDATION	32	0.0010	0.00000	0.2	(64,64,64) HU [-1350,150]	0.690	0.668	0.676	0.707	0.684
TEST	32	0.0010	0.00000	0.3	(64,64,64) HU [-1350,150]	0.656	0.679	0.783	0.544	0.643
VALIDATION	32	0.0010	0.00000	0.3	(64,64,64) HU [-1350,150]	0.630	0.656	0.760	0.520	0.622
TEST	4	0.0010	0.00000	0.2	(64,64,64) HU [-1350,150]	0.640	0.659	0.750	0.546	0.636
VALIDATION	4	0.0010	0.00000	0.2	(64,64,64) HU [-1350,150]	0.680	0.673	0.716	0.647	0.677
TEST	8	0.0010	0.00001	0.4	(64,64,64) HU [-1000,400]	0.640	0.674	0.800	0.501	0.629
VALIDATION	8	0.0010	0.00001	0.4	(64,64,64) HU [-1000,400]	0.670	0.664	0.720	0.627	0.663
TEST	8	0.0010	0.00000	0.4	(64,64,64) HU [-1350,150]	0.656	0.710	0.867	0.465	0.626
VALIDATION	8	0.0010	0.00000	0.4	(64,64,64) HU [-1350,150]	0.680	0.688	0.762	0.616	0.674
TEST	4	0.0010	0.00001	0.5	(128,128,128) HU [-1000,400]	0.648	0.671	0.783	0.534	0.619
VALIDATION	4	0.0010	0.00001	0.5	(128,128,128) HU [-1000,400]	0.650	0.655	0.736	0.571	0.634
TEST	4	0.0010	0.00000	0.4	(64,64,64) HU [-1350,150]	0.616	0.643	0.730	0.511	0.601
VALIDATION	4	0.0010	0.00000	0.4	(64,64,64) HU [-1350,150]	0.690	0.703	0.784	0.616	0.683
TEST	4	0.0010	0.00000	0.4	(128,256,256) HU [-1350,150]	0.600	0.641	0.761	0.452	0.572
VALIDATION	4	0.0010	0.00000	0.4	(128,256,256) HU [-1350,150]	0.670	0.669	0.720	0.635	0.668
TEST	8	0.0010	0.00000	0.4	(128,256,256) HU [-1350,150]	0.600	0.646	0.798	0.424	0.558
VALIDATION	8	0.0010	0.00000	0.4	(128,256,256) HU [-1350,150]	0.690	0.709	0.804	0.595	0.679
TEST	2	0.0001	0.00000	0.2	(128,256,256) HU [-1350,150]	0.576	0.605	0.714	0.453	0.541
VALIDATION	2	0.0001	0.00000	0.2	(128,256,256) HU [-1350,150]	0.660	0.677	0.784	0.558	0.656
TEST	8	0.0010	0.00000	0.2	(128,256,256) HU [-1000,400]	0.592	0.606	0.697	0.501	0.539
VALIDATION	8	0.0010	0.00000	0.2	(128,256,256) HU [-1000,400]	0.660	0.641	0.651	0.664	0.642
TEST	2	0.0001	0.00001	0.5	(128,128,128) HU [-1000,400]	0.536	0.527	0.580	0.501	0.522
VALIDATION	2	0.0001	0.00001	0.5	(128,128,128) HU [-1000,400]	0.600	0.570	0.602	0.591	0.587
TEST	4	0.0010	0.00001	0.5	(128,128,128) HU [-1350,150]	0.544	0.527	0.571	0.515	0.519
VALIDATION	4	0.0010	0.00001	0.5	(128,128,128) HU [-1350,150]	0.640	0.637	0.698	0.593	0.641

Volver arriba

Resultados DL3D multimodal

Resultados con Deep Learning con volúmenes 3D y datos clínicos (modelo multimodal) con el modelo DenseNet121

Set	Batch Size	LR	Dropout	Preprocesado	Accuracy	F1	TPR	TNR	G-Mean
TEST	4	0.001	0.4	(64,64,64) HU [-1000,400]	0.512	0.459	0.456	0.559	0.469
VALIDATION	4	0.001	0.4	(64,64,64) HU [-1000,400]	0.730	0.700	0.700	0.760	0.721
TEST	4	0.001	0.4	(64,64,64) HU [-1350,150]	0.488	0.460	0.458	0.515	0.448
VALIDATION	4	0.001	0.4	(64,64,64) HU [-1350,150]	0.710	0.686	0.700	0.724	0.706
TEST	16	0.001	0.4	(64,64,64) HU [-1000,400]	0.480	0.545	0.700	0.288	0.428
VALIDATION	16	0.001	0.4	(64,64,64) HU [-1000,400]	0.660	0.679	0.784	0.556	0.618
TEST	16	0.001	0.4	(64,64,64) HU [-1350,150]	0.496	0.552	0.662	0.344	0.413
VALIDATION	16	0.001	0.4	(64,64,64) HU [-1350,150]	0.650	0.685	0.807	0.520	0.561
TEST	4	0.0001	0.4	(64,64,64) HU [-1350,150]	0.552	0.632	0.833	0.297	0.347
VALIDATION	4	0.0001	0.4	(64,64,64) HU [-1350,150]	0.570	0.651	0.871	0.316	0.369
TEST	4	0.0001	0.4	(64,64,64) HU [-1000,400]	0.504	0.565	0.761	0.268	0.259
VALIDATION	4	0.0001	0.4	(64,64,64) HU [-1000,400]	0.500	0.550	0.722	0.318	0.340
TEST	16	0.0001	0.4	(64,64,64) HU [-1350,150]	0.520	0.560	0.818	0.231	0.139
VALIDATION	16	0.0001	0.4	(64,64,64) HU [-1350,150]	0.480	0.506	0.800	0.236	0.085
TEST	16	0.0001	0.4	(64,64,64) HU [-1000,400]	0.488	0.515	0.800	0.214	0.053
VALIDATION	16	0.0001	0.4	(64,64,64) HU [-1000,400]	0.510	0.521	0.800	0.260	0.110

Volver arriba

Resultados DL3D Fine-tuning

Resultados en test con validación cruzada y distintas funciones de pérdida

Set	Dataset	Modelo	Fine Tune	Batch Size	LR	Weight Decay	Dropout	Pretrained Size	Preprocesado	Loss	Stop Criterion	Accuracy	F1	TPR	TNR	G-Mean
TEST	nodulemnist3d	ResNet3D	descongelado	16	1e-4	1e-5	0	(64,64,64)	(64,64,64) HU[-1000,400]	CrossEntropyLoss+ContrastiveLoss	pérdida	0.5920	0.5995	0.6637	0.5330	0.5792
TEST	nodulemnist3d	DenseNet121	descongelado	16	1e-4	1e-5	0	(64,64,64)	(64,64,64) HU[-1000,400]	CrossEntropyLoss	pérdida	0.5840	0.5810	0.6288	0.5462	0.5792
TEST	nodulemnist3d	ResNet3D	descongelado	16	1e-4	1e-5	0	(64,64,64)	(64,64,64) HU[-1000,400]	CrossEntropyLoss+TripletLoss	pérdida	0.5920	0.5956	0.6788	0.5176	0.5695
TEST	organmnist3d	DenseNet121	congelado	16	1e-4	1e-5	0.3	(64,64,64)	(64,64,64) HU[-1000,400]	CrossEntropyLoss+TripletLoss	accuracy	0.5840	0.5990	0.6621	0.5143	0.5660

Volver arriba

Resultados con solo datos clínicos con AA clásico

KNN

n_neighbors	weights	metric	Accuracy	F1	TPR	TNR	G-Mean
5	uniform	euclidean	0.520	0.497	0.509	0.533	0.509
3	uniform	euclidean	0.512	0.476	0.474	0.547	0.503
7	uniform	manhattan	0.512	0.468	0.458	0.563	0.503
5	distance	euclidean	0.480	0.466	0.476	0.487	0.478
9	distance	manhattan	0.480	0.404	0.374	0.577	0.452

Random Forest

n_estimators	max_depth	min_samples_split	Accuracy	F1	TPR	TNR	G-Mean
100	5	2	0.552	0.443	0.373	0.712	0.515
200	10	4	0.536	0.442	0.392	0.665	0.508
300	None	2	0.504	0.443	0.424	0.576	0.485

LightGBM

n_estimators	max_depth	learning_rate	Accuracy	F1	TPR	TNR	G-Mean
200	6	0.05	0.552	0.473	0.426	0.667	0.531
100	4	0.1	0.544	0.477	0.442	0.637	0.528
300	8	0.03	0.544	0.468	0.426	0.653	0.524

XGBoost

n_estimators	max_depth	learning_rate	Accuracy	F1	TPR	TNR	G-Mean
400	6	0.01	0.576	0.498	0.444	0.695	0.553
200	5	0.05	0.568	0.503	0.461	0.664	0.552
350	12	0.02	0.568	0.494	0.444	0.679	0.548
300	8	0.02	0.568	0.494	0.444	0.679	0.548
300	8	0.03	0.560	0.489	0.444	0.679	0.547
150	8	0.07	0.560	0.489	0.442	0.664	0.540
250	10	0.05	0.560	0.479	0.426	0.679	0.534
100	4	0.1	0.552	0.480	0.444	0.651	0.534
100	3	0.2	0.536	0.455	0.408	0.649	0.512

Volver arriba

Resultados radiómica combinado con DML y datos clínicos

Radiómica + sin datos clínicos + sin DML

Modelo	Parámetros	Scaler	Preprocesado	Reducción de dimensionalidad	Accuracy	F1	TPR	TNR	G-Mean
KNN	n_neighbors=5, weights=distance	StandardScaler	(128,128,128), HU [-1350,150]	-	0.744	0.713	0.724	0.765	0.739
LightGBM	n_estimators=200, learning_rate=0.1, num_leaves=31	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.744	0.703	0.680	0.791	0.728
KNN	n_neighbors=5, weights=distance	-	(128,128,128), HU [-1350,150]	PCA (99%)	0.736	0.692	0.688	0.779	0.724
KNN	n_neighbors=7, weights=distance	MinMaxScaler	(128,256,256), HU [-1350,150]	-	0.728	0.711	0.688	0.766	0.719
KNN	n_neighbors=5, weights=uniform	StandardScaler	(128,128,128), HU [-1350,150]	-	0.720	0.679	0.724	0.721	0.716
KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1350,150]	Variance Threshold (0.01)	0.712	0.702	0.741	0.692	0.711
KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1350,150]	-	0.712	0.702	0.741	0.692	0.711
KNN	n_neighbors=7, weights=distance	MinMaxScaler	(128,128,128), HU [-1000,400]	PCA (90%)	0.712	0.706	0.756	0.674	0.709
LightGBM	n_estimators=100, learning_rate=0.1, num_leaves=50	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.728	0.679	0.645	0.791	0.708
LightGBM	n_estimators=50, learning_rate=0.1, num_leaves=50	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.720	0.676	0.647	0.777	0.705
SVM	C=10.0, kernel=rbf, gamma=scale	-	(128,256,256), HU [-1000,400]	PCA (99%)	0.712	0.683	0.688	0.735	0.704
LightGBM	n_estimators=100, learning_rate=0.1, num_leaves=50	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.712	0.681	0.682	0.733	0.703
SVM	C=10.0, kernel=rbf, gamma=scale	-	(128,256,256), HU [-1000,400]	PCA (95%)	0.712	0.683	0.688	0.734	0.702
KNN	n_neighbors=5, weights=uniform	-	(128,128,128), HU [-1350,150]	PCA (99%)	0.712	0.673	0.688	0.735	0.702
KNN	n_neighbors=3, weights=distance	StandardScaler	(128,128,128), HU [-1000,400]	Variance Threshold (0.01)	0.712	0.679	0.683	0.736	0.702
KNN	n_neighbors=7, weights=distance	-	(128,256,256), HU [-1350,150]	PCA (99%)	0.704	0.678	0.706	0.707	0.701
KNN	n_neighbors=7, weights=distance	StandardScaler	(128,128,128), HU [-1000,400]	PCA (95%)	0.704	0.696	0.753	0.662	0.701
DecisionTree	max_depth=5, min_samples_split=2	StandardScaler	(128,128,128), HU [-1350,150]	PCA (95%)	0.704	0.677	0.686	0.720	0.700
KNN	n_neighbors=7, weights=distance	MinMaxScaler	(128,128,128), HU [-1350,150]	-	0.712	0.664	0.653	0.765	0.700
KNN	n_neighbors=5, weights=distance	StandardScaler	(128,256,256), HU [-1000,400]	Variance Threshold (0.01)	0.712	0.674	0.668	0.748	0.700
KNN	n_neighbors=5, weights=distance	MinMaxScaler	(128,128,128), HU [-1350,150]	-	0.712	0.672	0.689	0.737	0.699
KNN	n_neighbors=7, weights=distance	StandardScaler	(128,128,128), HU [-1000,400]	PCA (90%)	0.704	0.703	0.774	0.645	0.699
KNN	n_neighbors=5, weights=distance	-	(128,128,128), HU [-1350,150]	PCA (95%)	0.704	0.669	0.670	0.735	0.698
LightGBM	n_estimators=50, learning_rate=0.1, num_leaves=31	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.682	0.698	0.704	0.698
LightGBM	n_estimators=200, learning_rate=0.1, num_leaves=31	MinMaxScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.712	0.675	0.679	0.733	0.698
KNN	n_neighbors=5, weights=distance	StandardScaler	(128,128,128), HU [-1350,150]	Variance Threshold (0.01)	0.712	0.670	0.667	0.752	0.698
KNN	n_neighbors=5, weights=distance	MinMaxScaler	(128,128,128), HU [-1000,400]	PCA (90%)	0.704	0.675	0.667	0.733	0.698
KNN	n_neighbors=7, weights=distance	StandardScaler	(128,256,256), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.665	0.671	0.734	0.696
SVM	C=10.0, kernel=rbf, gamma=scale	-	(128,256,256), HU [-1000,400]	PCA (90%)	0.696	0.687	0.738	0.662	0.695
KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1000,400]	-	0.696	0.686	0.723	0.678	0.695
KNN	n_neighbors=5, weights=distance	-	(128,128,128), HU [-1350,150]	PCA (90%)	0.704	0.664	0.670	0.735	0.694
KNN	n_neighbors=5, weights=distance	MinMaxScaler	(128,256,256), HU [-1000,400]	PCA (90%)	0.696	0.673	0.683	0.705	0.693
KNN	n_neighbors=5, weights=distance	StandardScaler	(128,128,128), HU [-1000,400]	-	0.696	0.670	0.686	0.704	0.693
RandomForest	n_estimators=50, max_depth=5, min_samples_split=5	MinMaxScaler	(128,128,128), HU [-1000,400]	PCA (90%)	0.704	0.668	0.633	0.763	0.692
KNN	n_neighbors=5, weights=uniform	StandardScaler	(128,128,128), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.663	0.667	0.737	0.692
XGBoost	n_estimators=100, learning_rate=0.1, max_depth=5	MinMaxScaler	(128,128,128), HU [-1000,400]	PCA (90%)	0.704	0.667	0.667	0.736	0.692
XGBoost	n_estimators=100, learning_rate=0.1, max_depth=5	-	(28,28,28), HU [-1350,150]	-	0.704	0.655	0.614	0.778	0.691
KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1350,150]	PCA (99%)	0.712	0.648	0.618	0.793	0.690
LightGBM	n_estimators=100, learning_rate=0.01, num_leaves=50	MinMaxScaler	(128,128,128), HU [-1350,150]	PCA (95%)	0.712	0.657	0.611	0.791	0.690

Radiómica + con datos clínicos + sin DML

Características	Modelo	Parámetros	Scaler	Preprocesado	Reducción de dimensionalidad	Accuracy	F1	TPR	TNR	G-Mean
Extendidas	LightGBM	n_estimators=100, learning_rate=0.1, num_leaves=31	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.760	0.725	0.700	0.805	0.746
Originales	GradientBoosting	n_estimators=200, learning_rate=0.01, max_depth=5	StandardScaler	(256,512,512), HU [-1000,400]	PCA (90%)	0.752	0.709	0.665	0.821	0.736
Extendidas	KNN	n_neighbors=5, weights=distance	MinMaxScaler	(128,128,128), HU [-1350,150]	-	0.728	0.684	0.689	0.766	0.712
Extendidas	KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1350,150]	-	0.712	0.702	0.741	0.692	0.711
Originales	GradientBoosting	n_estimators=100, learning_rate=0.01, max_depth=5	-	(256,512,512), HU [-1000,400]	PCA (90%)	0.728	0.673	0.614	0.822	0.709
Originales	XGBoost	n_estimators=200, learning_rate=0.1, max_depth=null	MinMaxScaler	(256,512,512), HU [-1350,150]	-	0.728	0.674	0.632	0.808	0.707
Extendidas	LightGBM	n_estimators=200, learning_rate=0.1, num_leaves=50	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.736	0.678	0.644	0.805	0.707
Extendidas	LightGBM	n_estimators=50, learning_rate=0.1, num_leaves=31	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.720	0.676	0.647	0.777	0.705
Extendidas	KNN	n_neighbors=5, weights=uniform	MinMaxScaler	(128,128,128), HU [-1350,150]	-	0.720	0.678	0.689	0.752	0.705
Originales	XGBoost	n_estimators=100, learning_rate=0.1, max_depth=null	-	(256,512,512), HU [-1350,150]	-	0.720	0.666	0.632	0.793	0.701
Extendidas	RandomForest	n_estimators=200, max_depth=10, min_samples_split=2	-	(128,128,128), HU [-1000,400]	PCA (90%)	0.720	0.667	0.612	0.808	0.700
Originales	XGBoost	n_estimators=200, learning_rate=0.1, max_depth=5	-	(256,512,512), HU [-1350,150]	-	0.712	0.667	0.650	0.764	0.699
Originales	GradientBoosting	n_estimators=100, learning_rate=0.01, max_depth=5	-	(256,512,512), HU [-1350,150]	-	0.720	0.666	0.598	0.820	0.699
Extendidas	LightGBM	n_estimators=200, learning_rate=0.1, num_leaves=50	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.720	0.671	0.661	0.762	0.697
Originales	SVM	C=1.0, kernel=rbf, gamma=scale	StandardScaler	(128,128,128), HU [-1350,150]	Variance Threshold (0.01)	0.720	0.663	0.615	0.808	0.697
Originales	GradientBoosting	n_estimators=200, learning_rate=0.01, max_depth=5	-	(256,512,512), HU [-1350,150]	-	0.712	0.661	0.615	0.791	0.696
Extendidas	RandomForest	n_estimators=100, max_depth=10, min_samples_split=5	-	(128,128,128), HU [-1000,400]	PCA (90%)	0.720	0.659	0.595	0.822	0.695
Originales	GradientBoosting	n_estimators=50, learning_rate=0.01, max_depth=null	StandardScaler	(256,512,512), HU [-1000,400]	PCA (90%)	0.712	0.662	0.614	0.793	0.695
Extendidas	LightGBM	n_estimators=50, learning_rate=0.1, num_leaves=50	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.673	0.680	0.719	0.695
Extendidas	KNN	n_neighbors=7, weights=distance	-	(128,128,128), HU [-1000,400]	-	0.696	0.686	0.723	0.678	0.695
Originales	XGBoost	n_estimators=50, learning_rate=0.1, max_depth=10	-	(256,512,512), HU [-1000,400]	PCA (95%)	0.712	0.667	0.645	0.762	0.694
Originales	XGBoost	n_estimators=50, learning_rate=0.1, max_depth=5	-	(256,512,512), HU [-1350,150]	-	0.712	0.658	0.614	0.793	0.694
Originales	GradientBoosting	n_estimators=200, learning_rate=0.1, max_depth=10	-	(256,512,512), HU [-1000,400]	PCA (90%)	0.704	0.665	0.648	0.748	0.692
Originales	XGBoost	n_estimators=100, learning_rate=0.1, max_depth=5	StandardScaler	(256,512,512), HU [-1350,150]	-	0.712	0.656	0.614	0.792	0.691
Originales	GradientBoosting	n_estimators=50, learning_rate=0.01, max_depth=5	StandardScaler	(256,512,512), HU [-1000,400]	PCA (90%)	0.712	0.654	0.595	0.807	0.691
Extendidas	XGBoost	n_estimators=100, learning_rate=0.1, max_depth=10	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.662	0.630	0.765	0.691
Originales	XGBoost	n_estimators=50, learning_rate=0.1, max_depth=10	StandardScaler	(256,512,512), HU [-1350,150]	-	0.712	0.652	0.614	0.793	0.690
Extendidas	LightGBM	n_estimators=100, learning_rate=0.1, num_leaves=31	-	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.665	0.662	0.733	0.690
Originales	XGBoost	n_estimators=200, learning_rate=0.1, max_depth=5	StandardScaler	(256,512,512), HU [-1000,400]	PCA (95%)	0.712	0.655	0.611	0.791	0.689
Originales	GradientBoosting	n_estimators=50, learning_rate=0.01, max_depth=5	-	(256,512,512), HU [-1350,150]	-	0.704	0.659	0.630	0.763	0.689
Extendidas	XGBoost	n_estimators=200, learning_rate=0.1, max_depth=10	StandardScaler	(28,28,28), HU [-1350,150]	Variance Threshold (0.01)	0.704	0.658	0.629	0.764	0.689
Extendidas	XGBoost	n_estimators=50, learning_rate=0.01, max_depth=10	MinMaxScaler	(128,128,128), HU [-1350,150]	PCA (95%)	0.704	0.658	0.629	0.764	0.689

Radiómica + sin datos clínicos + con DML

Número de vecinos	Weights	Scaler	Preprocesado	Algoritmo DML	Dimensión DML	Accuracy	F1	TPR	TNR	G-Mean
5	distance	StandardScaler	(128,128,128), HU [-1350,150]	nca	all	0.744	0.713	0.724	0.765	0.739
5	distance	-	(128,128,128), HU [-1350,150]	nca	20	0.736	0.707	0.723	0.749	0.732
5	distance	-	(128,128,128), multiwindowing	lmnn	all	0.733	0.704	0.727	0.738	0.731
7	distance	StandardScaler	(64,64,64), no HU	lmnn	all	0.747	0.696	0.667	0.810	0.725
5	uniform	MinMaxScaler	(64,64,64), no HU	lmnn	10	0.728	0.697	0.688	0.766	0.721
5	distance	-	(64,64,64), no HU	lmnn	20	0.752	0.686	0.598	0.880	0.721
5	distance	MinMaxScaler	(64,64,64), no HU	lmnn	10	0.728	0.691	0.670	0.780	0.720
3	distance	-	(128,128,128), HU [-1350,150]	lmnn	all	0.720	0.705	0.758	0.690	0.720
7	distance	MinMaxScaler	(128,256,256), HU [-1350,150]	nca	all	0.728	0.691	0.688	0.766	0.719
3	uniform	StandardScaler	(64,64,64), no HU	lmnn	all	0.733	0.676	0.636	0.810	0.717
3	distance	MinMaxScaler	(64,64,64), no HU	lmnn	20	0.720	0.692	0.691	0.746	0.717
3	uniform	MinMaxScaler	(64,64,64), no HU	lmnn	20	0.720	0.692	0.691	0.746	0.717
7	distance	StandardScaler	(64,64,64), no HU	lmnn	all	0.720	0.692	0.689	0.747	0.716
5	uniform	StandardScaler	(128,128,128), HU [-1350,150]	nca	all	0.720	0.694	0.724	0.721	0.716
7	distance	-	(128,128,128), HU [-1350,150]	nca	30	0.712	0.702	0.741	0.692	0.711
5	distance	-	(128,256,256), HU [-1350,150]	nca	10	0.712	0.691	0.702	0.723	0.710
7	uniform	StandardScaler	(64,64,64), no HU	lmnn	all	0.712	0.685	0.689	0.733	0.710
3	distance	MinMaxScaler	(128,128,128), HU [-1350,150]	lmnn	30	0.710	0.686	0.710	0.710	0.709
5	uniform	-	(128,128,128), HU [-1350,150]	nca	20	0.712	0.688	0.723	0.705	0.709
5	uniform	StandardScaler	(64,64,64), no HU	lmnn	30	0.712	0.688	0.689	0.732	0.708
5	distance	MinMaxScaler	(64,64,64), multiwindowing	lmnn	20	0.720	0.678	0.646	0.779	0.708
3	uniform	-	(64,64,64), no HU	lmnn	30	0.712	0.683	0.691	0.733	0.708
3	uniform	MinMaxScaler	(64,64,64), no HU	lmnn	30	0.712	0.688	0.689	0.732	0.707
5	distance	-	(128,128,128), HU [-1350,150]	nca	20	0.720	0.676	0.689	0.749	0.707
7	distance	MinMaxScaler	(64,64,64), no HU	lmnn	all	0.733	0.673	0.636	0.810	0.707
7	distance	StandardScaler	(128,256,256), HU [-1000,400]	nca	all	0.704	0.694	0.738	0.678	0.705
7	uniform	MinMaxScaler	(64,64,64), no HU	lmnn	30	0.712	0.681	0.671	0.747	0.705
5	distance	MinMaxScaler	(128,128,128), no HU	lmnn	20	0.712	0.682	0.668	0.748	0.705
5	uniform	-	(64,64,64), no HU	lmnn	20	0.736	0.669	0.582	0.865	0.704
5	distance	StandardScaler	(64,64,64), multiwindowing	lmnn	20	0.720	0.672	0.642	0.780	0.704
3	distance	-	(128,128,128), multiwindowing	lmnn	all	0.707	0.685	0.727	0.690	0.704
7	distance	-	(128,128,128), HU [-1000,400]	nca	10	0.704	0.690	0.723	0.692	0.704
7	distance	-	(128,128,128), HU [-1350,150]	nca	10	0.704	0.696	0.741	0.678	0.703
5	uniform	-	(128,128,128), multiwindowing	lmnn	all	0.707	0.672	0.697	0.714	0.703
5	uniform	-	(128,256,256), HU [-1350,150]	nca	10	0.704	0.691	0.720	0.693	0.703
3	distance	MinMaxScaler	(64,64,64), no HU	lmnn	20	0.704	0.679	0.691	0.718	0.702
7	distance	StandardScaler	(128,256,256), HU [-1350,150]	nca	20	0.712	0.675	0.653	0.764	0.702
7	distance	-	(128,128,128), HU [-1350,150]	nca	20	0.720	0.665	0.655	0.779	0.702
7	distance	MinMaxScaler	(64,64,64), no HU	lmnn	all	0.704	0.681	0.689	0.718	0.701
7	distance	StandardScaler	(128,128,128), HU [-1000,400]	lmnn	10	0.710	0.681	0.695	0.727	0.701
5	distance	StandardScaler	(128,128,128), multiwindowing	nca	all	0.712	0.680	0.685	0.735	0.701
5	distance	MinMaxScaler	(128,128,128), multiwindowing	nca	all	0.712	0.681	0.683	0.734	0.701

Radiómica + con datos clínicos + con DML

Número de vecinos	Weights	Scaler	Preprocesado	Algoritmo DML	Dimensión DML	Accuracy	F1	TPR	TNR	G-Mean
5	distance	-	(128,128,128), multiwindowing	lmnn	20	0.747	0.727	0.758	0.738	0.747
5	distance	StandardScaler	(128,128,128), multiwindowing	lmnn	20	0.747	0.708	0.697	0.786	0.739
5	uniform	-	(128,128,128), multiwindowing	lmnn	20	0.733	0.715	0.758	0.714	0.735
7	distance	MinMaxScaler	(128,128,128), multiwindowing	lmnn	20	0.733	0.708	0.727	0.738	0.733
7	distance	-	(128,128,128), multiwindowing	lmnn	20	0.733	0.708	0.727	0.738	0.731
5	distance	-	(64,64,64), no HU	lmnn	20	0.760	0.700	0.623	0.870	0.731
5	distance	-	(128,128,128), multiwindowing	lmnn	all	0.733	0.704	0.727	0.738	0.731
3	uniform	MinMaxScaler	(64,64,64), no HU	lmnn	20	0.750	0.693	0.623	0.850	0.724
7	uniform	-	(128,128,128), multiwindowing	lmnn	20	0.720	0.696	0.727	0.714	0.721
5	distance	-	(64,64,64), no HU	lmnn	20	0.752	0.686	0.598	0.880	0.721
7	uniform	-	(128,128,128), multiwindowing	lmnn	20	0.720	0.697	0.727	0.714	0.720
3	distance	-	(128,128,128), HU [-1350,150]	lmnn	all	0.720	0.705	0.758	0.690	0.720
7	distance	StandardScaler	(128,128,128), HU [-1000,400]	nca	20	0.736	0.686	0.652	0.807	0.718
5	uniform	StandardScaler	(128,128,128), multiwindowing	lmnn	20	0.720	0.686	0.697	0.738	0.717
5	distance	MinMaxScaler	(128,128,128), HU [-1350,150]	nca	all	0.728	0.684	0.689	0.766	0.712
7	distance	-	(128,128,128), HU [-1350,150]	nca	30	0.712	0.702	0.741	0.692	0.711
5	uniform	-	(64,64,64), no HU	lmnn	20	0.740	0.678	0.602	0.850	0.711
3	distance	StandardScaler	(128,128,128), multiwindowing	lmnn	20	0.720	0.678	0.667	0.762	0.710
7	distance	MinMaxScaler	(128,128,128), multiwindowing	lmnn	all	0.720	0.686	0.697	0.738	0.710
3	distance	MinMaxScaler	(128,128,128), HU [-1350,150]	lmnn	30	0.710	0.686	0.710	0.710	0.709
7	distance	StandardScaler	(128,128,128), HU [-1000,400]	nca	20	0.728	0.678	0.652	0.795	0.708
7	distance	StandardScaler	(128,128,128), HU [-1000,400]	lmnn	all	0.707	0.686	0.727	0.690	0.708
3	uniform	StandardScaler	(64,64,64), no HU	lmnn	20	0.740	0.674	0.600	0.850	0.708
7	distance	StandardScaler	(128,128,128), multiwindowing	lmnn	20	0.707	0.688	0.727	0.690	0.707
7	distance	StandardScaler	(64,64,64), no HU	lmnn	all	0.733	0.673	0.636	0.810	0.707
5	uniform	MinMaxScaler	(128,128,128), HU [-1350,150]	nca	all	0.720	0.678	0.689	0.752	0.705
7	uniform	StandardScaler	(64,64,64), no HU	lmnn	20	0.712	0.677	0.670	0.749	0.705
7	distance	StandardScaler	(64,64,64), no HU	lmnn	20	0.712	0.677	0.670	0.749	0.705
5	uniform	-	(64,64,64), no HU	lmnn	20	0.736	0.669	0.582	0.865	0.704
3	distance	-	(128,128,128), multiwindowing	lmnn	all	0.707	0.685	0.727	0.690	0.704
7	distance	-	(128,128,128), HU [-1350,150]	nca	10	0.704	0.696	0.741	0.678	0.703
7	distance	StandardScaler	(128,256,256), HU [-1350,150]	nca	20	0.704	0.687	0.705	0.704	0.703
5	uniform	-	(128,128,128), multiwindowing	lmnn	all	0.707	0.672	0.697	0.714	0.703
7	distance	StandardScaler	(128,128,128), multiwindowing	lmnn	all	0.707	0.679	0.697	0.714	0.701
3	distance	MinMaxScaler	(64,64,64), multiwindowing	lmnn	20	0.700	0.681	0.714	0.690	0.701
7	distance	StandardScaler	(128,128,128), HU [-1000,400]	lmnn	10	0.710	0.681	0.695	0.727	0.701
7	distance	MinMaxScaler	(128,128,128), HU [-1350,150]	nca	all	0.712	0.667	0.670	0.751	0.701
5	distance	MinMaxScaler	(128,128,128), HU [-1350,150]	nca	all	0.712	0.672	0.689	0.736	0.700
3	distance	MinMaxScaler	(64,64,64), no HU	lmnn	all	0.720	0.656	0.606	0.810	0.700