Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection

Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection. Wu, Y., Schmidt, A., Hernández-Sánchez, E., Molina, R., & Katsaggelos, A. K. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), volume 12902 LNCS, pages 582–591, 2021.

Paper doi abstract bibtex

Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.

@inproceedings{Yunan2021a,
abstract = {Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.},
author = {Wu, Yunan and Schmidt, Arne and Hern{\'{a}}ndez-S{\'{a}}nchez, Enrique and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
doi = {10.1007/978-3-030-87196-3_54},
isbn = {9783030871956},
issn = {16113349},
keywords = {Attention-based multiple instance learning,CT hemorrhage detection,Variational Gaussian processes},
pages = {582--591},
title = {{Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection}},
url = {https://link.springer.com/10.1007/978-3-030-87196-3_54},
volume = {12902 LNCS},
year = {2021}
}

Downloads: 0

{"_id":"3AuJpcTQYssCacdM9","bibbaseid":"wu-schmidt-hernndezsnchez-molina-katsaggelos-combiningattentionbasedmultipleinstancelearningandgaussianprocessesforcthemorrhagedetection-2021","author_short":["Wu, Y.","Schmidt, A.","Hernández-Sánchez, E.","Molina, R.","Katsaggelos, A. K."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.","author":[{"propositions":[],"lastnames":["Wu"],"firstnames":["Yunan"],"suffixes":[]},{"propositions":[],"lastnames":["Schmidt"],"firstnames":["Arne"],"suffixes":[]},{"propositions":[],"lastnames":["Hernández-Sánchez"],"firstnames":["Enrique"],"suffixes":[]},{"propositions":[],"lastnames":["Molina"],"firstnames":["Rafael"],"suffixes":[]},{"propositions":[],"lastnames":["Katsaggelos"],"firstnames":["Aggelos","K."],"suffixes":[]}],"booktitle":"Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)","doi":"10.1007/978-3-030-87196-3_54","isbn":"9783030871956","issn":"16113349","keywords":"Attention-based multiple instance learning,CT hemorrhage detection,Variational Gaussian processes","pages":"582–591","title":"Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection","url":"https://link.springer.com/10.1007/978-3-030-87196-3_54","volume":"12902 LNCS","year":"2021","bibtex":"@inproceedings{Yunan2021a,\nabstract = {Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.},\nauthor = {Wu, Yunan and Schmidt, Arne and Hern{\\'{a}}ndez-S{\\'{a}}nchez, Enrique and Molina, Rafael and Katsaggelos, Aggelos K.},\nbooktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},\ndoi = {10.1007/978-3-030-87196-3_54},\nisbn = {9783030871956},\nissn = {16113349},\nkeywords = {Attention-based multiple instance learning,CT hemorrhage detection,Variational Gaussian processes},\npages = {582--591},\ntitle = {{Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection}},\nurl = {https://link.springer.com/10.1007/978-3-030-87196-3_54},\nvolume = {12902 LNCS},\nyear = {2021}\n}\n","author_short":["Wu, Y.","Schmidt, A.","Hernández-Sánchez, E.","Molina, R.","Katsaggelos, A. K."],"key":"Yunan2021a","id":"Yunan2021a","bibbaseid":"wu-schmidt-hernndezsnchez-molina-katsaggelos-combiningattentionbasedmultipleinstancelearningandgaussianprocessesforcthemorrhagedetection-2021","role":"author","urls":{"Paper":"https://link.springer.com/10.1007/978-3-030-87196-3_54"},"keyword":["Attention-based multiple instance learning","CT hemorrhage detection","Variational Gaussian processes"],"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://sites.northwestern.edu/ivpl/files/2023/06/IVPL_Updated_publications-1.bib","dataSources":["KTWAakbPXLGfYseXn","ePKPjG8C6yvpk4mEK","ya2CyA73rpZseyrZ8","zFPgsTDAW8aDnb5iN","E6Bth2QB5BYjBMZE7","nbnEjsN7MJhurAK9x","PNQZj6FjzoxxJk4Yi","7FpDWDGJ4KgpDiGfB","bod9ms4MQJHuJgPpp","QR9t5P2cLdJuzhfzK","D8k2SxfC5dKNRFgro","7Dwzbxq93HWrJEhT6","qhF8zxmGcJfvtdeAg","fvDEHD49E2ZRwE3fb","H7crv8NWhZup4d4by","DHqokWsryttGh7pJE","vRJd4wNg9HpoZSMHD","sYxQ6pxFgA59JRhxi","w2WahSbYrbcCKBDsC","XasdXLL99y5rygCmq","3gkSihZQRfAD2KBo3","t5XMbyZbtPBo4wBGS","bEpHM2CtrwW2qE8FP","teJzFLHexaz5AQW5z"],"keywords":["attention-based multiple instance learning","ct hemorrhage detection","variational gaussian processes"],"search_terms":["combining","attention","based","multiple","instance","learning","gaussian","processes","hemorrhage","detection","wu","schmidt","hernández-sánchez","molina","katsaggelos"],"title":"Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection","year":2021}