USE-Net: incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets

Leonardo Rundo (Corresponding author), Changee Han, Yudai Nagano, Jin Zhang, Ryuichiro Hataya, Carmelo Militello, Andrea Tangherloni, Marco Nobile, Claudio Ferretti, Daniela Besozzi, Maria Carla Gilardi, Salvatore Vitabile, Giancarlo Mauri, Hideki Nakayama, Paolo Cazzaniga

Onderzoeksoutput: Bijdrage aan tijdschriftTijdschriftartikelAcademicpeer review

1 Citaat (Scopus)

Uittreksel

Prostate cancer is the most common malignant tumors in men but prostate Magnetic Resonance Imaging (MRI) analysis remains challenging. Besides whole prostate gland segmentation, the capability to differentiate between the blurry boundary of the Central Gland (CG) and Peripheral Zone (PZ) can lead to differential diagnosis, since the frequency and severity of tumors differ in these regions. To tackle the prostate zonal segmentation task, we propose a novel Convolutional Neural Network (CNN), called USE-Net, which incorporates Squeeze-and-Excitation (SE) blocks into U-Net, i.e., one of the most effective CNNs in biomedical image segmentation. Especially, the SE blocks are added after every Encoder (Enc USE-Net) or Encoder-Decoder block (Enc-Dec USE-Net). This study evaluates the generalization ability of CNN-based architectures on three T2-weighted MRI datasets, each one consisting of a different number of patients and heterogeneous image characteristics, collected by different institutions. The following mixed scheme is used for training/testing: (i) training on either each individual dataset or multiple prostate MRI datasets and (ii) testing on all three datasets with all possible training/testing combinations. USE-Net is compared against three state-of-the-art CNN-based architectures (i.e., U-Net, pix2pix, and Mixed-Scale Dense Network), along with a semi-automatic continuous max-flow model. The results show that training on the union of the datasets generally outperforms training on each dataset separately, allowing for both intra-/cross-dataset generalization. Enc USE-Net shows good overall generalization under any training condition, while Enc-Dec USE-Net remarkably outperforms the other methods when trained on all datasets. These findings reveal that the SE blocks’ adaptive feature recalibration provides excellent cross-dataset generalization when testing is performed on samples of the datasets used during training. Therefore, we should consider multi-dataset training and SE blocks together as mutually indispensable methods to draw out each other’s full potential. In conclusion, adaptive mechanisms (e.g., feature recalibration) may be a valuable solution in medical imaging applications involving multi-institutional settings.
TaalEngels
Pagina's31-43
TijdschriftNeurocomputing
Volume365
DOI's
StatusGepubliceerd - 6 nov 2019
Extern gepubliceerdJa

Vingerafdruk

Prostate
Magnetic Resonance Imaging
Testing
Neural networks
Tumors
Medical imaging
Image segmentation
Datasets
Diagnostic Imaging
Neoplasms
Prostatic Neoplasms
Differential Diagnosis

Trefwoorden

    Citeer dit

    Rundo, L., Han, C., Nagano, Y., Zhang, J., Hataya, R., Militello, C., ... Cazzaniga, P. (2019). USE-Net: incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets. Neurocomputing, 365, 31-43. DOI: 10.1016/j.neucom.2019.07.006
    Rundo, Leonardo ; Han, Changee ; Nagano, Yudai ; Zhang, Jin ; Hataya, Ryuichiro ; Militello, Carmelo ; Tangherloni, Andrea ; Nobile, Marco ; Ferretti, Claudio ; Besozzi, Daniela ; Gilardi, Maria Carla ; Vitabile, Salvatore ; Mauri, Giancarlo ; Nakayama, Hideki ; Cazzaniga, Paolo. / USE-Net : incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets. In: Neurocomputing. 2019 ; Vol. 365. blz. 31-43
    @article{905713f45f564349abfa85d10e02655b,
    title = "USE-Net: incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets",
    abstract = "Prostate cancer is the most common malignant tumors in men but prostate Magnetic Resonance Imaging (MRI) analysis remains challenging. Besides whole prostate gland segmentation, the capability to differentiate between the blurry boundary of the Central Gland (CG) and Peripheral Zone (PZ) can lead to differential diagnosis, since the frequency and severity of tumors differ in these regions. To tackle the prostate zonal segmentation task, we propose a novel Convolutional Neural Network (CNN), called USE-Net, which incorporates Squeeze-and-Excitation (SE) blocks into U-Net, i.e., one of the most effective CNNs in biomedical image segmentation. Especially, the SE blocks are added after every Encoder (Enc USE-Net) or Encoder-Decoder block (Enc-Dec USE-Net). This study evaluates the generalization ability of CNN-based architectures on three T2-weighted MRI datasets, each one consisting of a different number of patients and heterogeneous image characteristics, collected by different institutions. The following mixed scheme is used for training/testing: (i) training on either each individual dataset or multiple prostate MRI datasets and (ii) testing on all three datasets with all possible training/testing combinations. USE-Net is compared against three state-of-the-art CNN-based architectures (i.e., U-Net, pix2pix, and Mixed-Scale Dense Network), along with a semi-automatic continuous max-flow model. The results show that training on the union of the datasets generally outperforms training on each dataset separately, allowing for both intra-/cross-dataset generalization. Enc USE-Net shows good overall generalization under any training condition, while Enc-Dec USE-Net remarkably outperforms the other methods when trained on all datasets. These findings reveal that the SE blocks’ adaptive feature recalibration provides excellent cross-dataset generalization when testing is performed on samples of the datasets used during training. Therefore, we should consider multi-dataset training and SE blocks together as mutually indispensable methods to draw out each other’s full potential. In conclusion, adaptive mechanisms (e.g., feature recalibration) may be a valuable solution in medical imaging applications involving multi-institutional settings.",
    keywords = "Prostate zonal segmentation, Prostate cancer, Anatomical MRI, Convolutional neural networks, USE-Net, Cross-dataset generalization",
    author = "Leonardo Rundo and Changee Han and Yudai Nagano and Jin Zhang and Ryuichiro Hataya and Carmelo Militello and Andrea Tangherloni and Marco Nobile and Claudio Ferretti and Daniela Besozzi and Gilardi, {Maria Carla} and Salvatore Vitabile and Giancarlo Mauri and Hideki Nakayama and Paolo Cazzaniga",
    year = "2019",
    month = "11",
    day = "6",
    doi = "10.1016/j.neucom.2019.07.006",
    language = "English",
    volume = "365",
    pages = "31--43",
    journal = "Neurocomputing",
    issn = "0925-2312",
    publisher = "Elsevier",

    }

    Rundo, L, Han, C, Nagano, Y, Zhang, J, Hataya, R, Militello, C, Tangherloni, A, Nobile, M, Ferretti, C, Besozzi, D, Gilardi, MC, Vitabile, S, Mauri, G, Nakayama, H & Cazzaniga, P 2019, 'USE-Net: incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets' Neurocomputing, vol. 365, blz. 31-43. DOI: 10.1016/j.neucom.2019.07.006

    USE-Net : incorporating squeeze-and-excitation blocks into U-net for prostate zonal segmentation of multi-institutional MRI datasets. / Rundo, Leonardo (Corresponding author); Han, Changee; Nagano, Yudai; Zhang, Jin; Hataya, Ryuichiro; Militello, Carmelo; Tangherloni, Andrea; Nobile, Marco; Ferretti, Claudio; Besozzi, Daniela; Gilardi, Maria Carla; Vitabile, Salvatore; Mauri, Giancarlo; Nakayama, Hideki; Cazzaniga, Paolo.

    In: Neurocomputing, Vol. 365, 06.11.2019, blz. 31-43.

    Onderzoeksoutput: Bijdrage aan tijdschriftTijdschriftartikelAcademicpeer review

    TY - JOUR

    T1 - USE-Net

    T2 - Neurocomputing

    AU - Rundo,Leonardo

    AU - Han,Changee

    AU - Nagano,Yudai

    AU - Zhang,Jin

    AU - Hataya,Ryuichiro

    AU - Militello,Carmelo

    AU - Tangherloni,Andrea

    AU - Nobile,Marco

    AU - Ferretti,Claudio

    AU - Besozzi,Daniela

    AU - Gilardi,Maria Carla

    AU - Vitabile,Salvatore

    AU - Mauri,Giancarlo

    AU - Nakayama,Hideki

    AU - Cazzaniga,Paolo

    PY - 2019/11/6

    Y1 - 2019/11/6

    N2 - Prostate cancer is the most common malignant tumors in men but prostate Magnetic Resonance Imaging (MRI) analysis remains challenging. Besides whole prostate gland segmentation, the capability to differentiate between the blurry boundary of the Central Gland (CG) and Peripheral Zone (PZ) can lead to differential diagnosis, since the frequency and severity of tumors differ in these regions. To tackle the prostate zonal segmentation task, we propose a novel Convolutional Neural Network (CNN), called USE-Net, which incorporates Squeeze-and-Excitation (SE) blocks into U-Net, i.e., one of the most effective CNNs in biomedical image segmentation. Especially, the SE blocks are added after every Encoder (Enc USE-Net) or Encoder-Decoder block (Enc-Dec USE-Net). This study evaluates the generalization ability of CNN-based architectures on three T2-weighted MRI datasets, each one consisting of a different number of patients and heterogeneous image characteristics, collected by different institutions. The following mixed scheme is used for training/testing: (i) training on either each individual dataset or multiple prostate MRI datasets and (ii) testing on all three datasets with all possible training/testing combinations. USE-Net is compared against three state-of-the-art CNN-based architectures (i.e., U-Net, pix2pix, and Mixed-Scale Dense Network), along with a semi-automatic continuous max-flow model. The results show that training on the union of the datasets generally outperforms training on each dataset separately, allowing for both intra-/cross-dataset generalization. Enc USE-Net shows good overall generalization under any training condition, while Enc-Dec USE-Net remarkably outperforms the other methods when trained on all datasets. These findings reveal that the SE blocks’ adaptive feature recalibration provides excellent cross-dataset generalization when testing is performed on samples of the datasets used during training. Therefore, we should consider multi-dataset training and SE blocks together as mutually indispensable methods to draw out each other’s full potential. In conclusion, adaptive mechanisms (e.g., feature recalibration) may be a valuable solution in medical imaging applications involving multi-institutional settings.

    AB - Prostate cancer is the most common malignant tumors in men but prostate Magnetic Resonance Imaging (MRI) analysis remains challenging. Besides whole prostate gland segmentation, the capability to differentiate between the blurry boundary of the Central Gland (CG) and Peripheral Zone (PZ) can lead to differential diagnosis, since the frequency and severity of tumors differ in these regions. To tackle the prostate zonal segmentation task, we propose a novel Convolutional Neural Network (CNN), called USE-Net, which incorporates Squeeze-and-Excitation (SE) blocks into U-Net, i.e., one of the most effective CNNs in biomedical image segmentation. Especially, the SE blocks are added after every Encoder (Enc USE-Net) or Encoder-Decoder block (Enc-Dec USE-Net). This study evaluates the generalization ability of CNN-based architectures on three T2-weighted MRI datasets, each one consisting of a different number of patients and heterogeneous image characteristics, collected by different institutions. The following mixed scheme is used for training/testing: (i) training on either each individual dataset or multiple prostate MRI datasets and (ii) testing on all three datasets with all possible training/testing combinations. USE-Net is compared against three state-of-the-art CNN-based architectures (i.e., U-Net, pix2pix, and Mixed-Scale Dense Network), along with a semi-automatic continuous max-flow model. The results show that training on the union of the datasets generally outperforms training on each dataset separately, allowing for both intra-/cross-dataset generalization. Enc USE-Net shows good overall generalization under any training condition, while Enc-Dec USE-Net remarkably outperforms the other methods when trained on all datasets. These findings reveal that the SE blocks’ adaptive feature recalibration provides excellent cross-dataset generalization when testing is performed on samples of the datasets used during training. Therefore, we should consider multi-dataset training and SE blocks together as mutually indispensable methods to draw out each other’s full potential. In conclusion, adaptive mechanisms (e.g., feature recalibration) may be a valuable solution in medical imaging applications involving multi-institutional settings.

    KW - Prostate zonal segmentation

    KW - Prostate cancer

    KW - Anatomical MRI

    KW - Convolutional neural networks

    KW - USE-Net

    KW - Cross-dataset generalization

    U2 - 10.1016/j.neucom.2019.07.006

    DO - 10.1016/j.neucom.2019.07.006

    M3 - Article

    VL - 365

    SP - 31

    EP - 43

    JO - Neurocomputing

    JF - Neurocomputing

    SN - 0925-2312

    ER -