Table 2.

Multi-label weighted mean log loss on the official test set of the RSNA Intracranial Hemorrhage Detection challenge for different deep learning models. Results are reported for different CNN and LSTM architectures with and without feature selection. The Kaggle rank of each submission is also reported here. The best results are highlighted in bold.

Method	BiLSTM Model	Features	Feature Count	Loss	Rank on Kaggle
EfficientNet-B4	-	Full feature vectors	2048	0.07212	278
ResNeXt-101 $32\times 8$ d	-	Full feature vectors	2048	0.06259	191
ResNeXt-101 + BiLSTM	$3\times 256$	ResNeXt-101 predictions	6	0.05540	64
ResNeXt-101 + BiLSTM	$3\times 256$	Features with largest standard deviations	192	0.05212	37
ResNeXt-101 + BiLSTM	$3\times 256$	Features of weights with largest magnitude	192	0.05365	42
ResNeXt-101 + BiLSTM	$3\times 256$	Features of weights with smallest magnitude	192	0.05136	34
ResNeXt-101 + BiLSTM	$3\times 256$	Features of weights with smallest magnitude	120	0.05035	29
ResNeXt-101 + BiLSTM	$2\times 128$	PCA features	192	0.05207	36
ResNeXt-101 + BiLSTM	$3\times 128$	PCA features	192	0.05198	35
ResNeXt-101 + BiLSTM	$3\times 256$	PCA features	192	0.05096	30
ResNeXt-101 + BiLSTM	$3\times 256$	PCA features	120	0.05022	29
ResNeXt-101 + BiLSTM	$3\times 256$	PCA features + ResNeXt-101 predictions	120	0.04989	27