notesum.ai

Published at December 5

Automated Medical Report Generation for ECG Data: Bridging Medical Text and Signal Processing with Deep Learning

cs.CL

cs.AI

Released Date: December 5, 2024

Authors: Amnon Bleich¹, Antje Linnemann², Bjoern H. Diem, Tim OF Conrad¹

Aff.: ¹University of Berlin; ²BIOTRONIK SE & Co KG

Arxiv: http://arxiv.org/pdf/2412.04067v1

Refer to caption

Encoder	Decoder	Additional	METEOR	BLEU (%)			ROUGE-1 (%)
Encoder	Decoder	Additional	(%)	1	2	4	P	R	F
ResNet34	LSTM	Trans./Abbr.	55.53	51.63	44.54	35.29	61.47	60.65	58.33
ResNet34	LSTM	Trans.	55.01	51.03	43.72	34.39	61.76	59.65	57.93
ResNet34	LSTM	-	50.72	48.25	42.29	34.90	55.53	54.93	52.29
ResNet34	Transformer	Trans./Abbr.	55.00	51.41	44.04	34.62	61.74	59.82	58.12
ResNet34	Transformer	Trans.	54.84	50.00	42.76	33.64	62.74	59.04	58.18
ResNet34	Transformer	-	51.11	47.21	41.02	33.29	57.61	54.76	53.87
Pre. ResNet34	LSTM	Trans./Abbr.	53.19	48.75	41.46	32.26	59.54	57.90	55.82
Pre. ResNet34	Transformer	Trans./Abbr.	52.83	48.93	41.42	31.93	60.77	57.96	56.73
ResNet18	LSTM	Trans./Abbr.	55.30	51.40	44.39	35.37	62.40	59.57	58.33
ResNet18	Transformer	Trans./Abbr.	55.00	50.95	43.70	34.53	63.47	59.39	58.82
Reference (Qiu et al.)*			24.51	27.21	-	-	26.12	35.71	29.56