Explainable Model for Localization of Spiculation in Lung Nodules

Mirtha Lucas; Miguel Lerma; Jacob Furst; Daniela Raicu

doi:10.1007/978-3-031-25082-8_30

Explainable Model for Localization of Spiculation in Lung Nodules

Mirtha Lucas^*, Miguel Lerma, Jacob Furst, Daniela Raicu

^*Corresponding author for this work

Mathematics

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

When determining a lung nodule malignancy one must consider the spiculation represented by spike-like structures in the nodule’s boundary. In this paper, we develop a deep learning model based on a VGG16 architecture to locate the presence of spiculation in lung nodules from Computed Tomography images. In order to increase the expert’s confidence in the model output, we apply our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping attribution method to visualize areas of the image (spicules). Therefore, the attribution method is applied to the layer of the model that is responsible for the detection of the spiculation features. We show that the first layers of the network are specialized in detecting low-level features such as edges, the last convolutional layer detects the general area occupied by the nodule, and finally, we identify that spiculation structures are detected at an intermediate layer. We use three different metrics to support our findings.

Original language	English (US)
Title of host publication	Computer Vision – ECCV 2022 Workshops, Proceedings
Editors	Leonid Karlinsky, Tomer Michaeli, Ko Nishino
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	457-471
Number of pages	15
ISBN (Print)	9783031250811
DOIs	https://doi.org/10.1007/978-3-031-25082-8_30
State	Published - 2023
Event	17th European Conference on Computer Vision, ECCV 2022 - Tel Aviv, Israel Duration: Oct 23 2022 → Oct 27 2022

Publication series

Name	Lecture Notes in Computer Science
Volume	13807 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	17th European Conference on Computer Vision, ECCV 2022
Country/Territory	Israel
City	Tel Aviv
Period	10/23/22 → 10/27/22

Funding

The work performed here has been restricted to one network architecture (VGG16) performing binary classification, one image domain (images of lung nodules from the LICD-IDRI dataset), and one semantic feature (spiculation). Further work can be made to adapt the methods used here to other network architectures (e.g. ResNet, Siamese networks, etc.), multiclass classification (e.g. by adding the middle spiculation levels), data domains (e.g. natural language), and features (e.g. sample similarity).

Keywords

Artificial intelligence
CAD
Computer-aided detection
Imaging informatics
XAI

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-031-25082-8_30

Cite this

Lucas, M., Lerma, M., Furst, J., & Raicu, D. (2023). Explainable Model for Localization of Spiculation in Lung Nodules. In L. Karlinsky, T. Michaeli, & K. Nishino (Eds.), Computer Vision – ECCV 2022 Workshops, Proceedings (pp. 457-471). (Lecture Notes in Computer Science; Vol. 13807 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-25082-8_30

@inproceedings{ac4eff8963e8470f86e7caee9cd44b95,

title = "Explainable Model for Localization of Spiculation in Lung Nodules",

abstract = "When determining a lung nodule malignancy one must consider the spiculation represented by spike-like structures in the nodule{\textquoteright}s boundary. In this paper, we develop a deep learning model based on a VGG16 architecture to locate the presence of spiculation in lung nodules from Computed Tomography images. In order to increase the expert{\textquoteright}s confidence in the model output, we apply our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping attribution method to visualize areas of the image (spicules). Therefore, the attribution method is applied to the layer of the model that is responsible for the detection of the spiculation features. We show that the first layers of the network are specialized in detecting low-level features such as edges, the last convolutional layer detects the general area occupied by the nodule, and finally, we identify that spiculation structures are detected at an intermediate layer. We use three different metrics to support our findings.",

keywords = "Artificial intelligence, CAD, Computer-aided detection, Imaging informatics, XAI",

author = "Mirtha Lucas and Miguel Lerma and Jacob Furst and Daniela Raicu",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 17th European Conference on Computer Vision, ECCV 2022 ; Conference date: 23-10-2022 Through 27-10-2022",

year = "2023",

doi = "10.1007/978-3-031-25082-8_30",

language = "English (US)",

isbn = "9783031250811",

series = "Lecture Notes in Computer Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "457--471",

editor = "Leonid Karlinsky and Tomer Michaeli and Ko Nishino",

booktitle = "Computer Vision – ECCV 2022 Workshops, Proceedings",

address = "Germany",

}

Lucas, M, Lerma, M, Furst, J & Raicu, D 2023, Explainable Model for Localization of Spiculation in Lung Nodules. in L Karlinsky, T Michaeli & K Nishino (eds), Computer Vision – ECCV 2022 Workshops, Proceedings. Lecture Notes in Computer Science, vol. 13807 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 457-471, 17th European Conference on Computer Vision, ECCV 2022, Tel Aviv, Israel, 10/23/22. https://doi.org/10.1007/978-3-031-25082-8_30

Explainable Model for Localization of Spiculation in Lung Nodules. / Lucas, Mirtha; Lerma, Miguel; Furst, Jacob et al.
Computer Vision – ECCV 2022 Workshops, Proceedings. ed. / Leonid Karlinsky; Tomer Michaeli; Ko Nishino. Springer Science and Business Media Deutschland GmbH, 2023. p. 457-471 (Lecture Notes in Computer Science; Vol. 13807 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Explainable Model for Localization of Spiculation in Lung Nodules

AU - Lucas, Mirtha

AU - Lerma, Miguel

AU - Furst, Jacob

AU - Raicu, Daniela

PY - 2023

Y1 - 2023

N2 - When determining a lung nodule malignancy one must consider the spiculation represented by spike-like structures in the nodule’s boundary. In this paper, we develop a deep learning model based on a VGG16 architecture to locate the presence of spiculation in lung nodules from Computed Tomography images. In order to increase the expert’s confidence in the model output, we apply our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping attribution method to visualize areas of the image (spicules). Therefore, the attribution method is applied to the layer of the model that is responsible for the detection of the spiculation features. We show that the first layers of the network are specialized in detecting low-level features such as edges, the last convolutional layer detects the general area occupied by the nodule, and finally, we identify that spiculation structures are detected at an intermediate layer. We use three different metrics to support our findings.

AB - When determining a lung nodule malignancy one must consider the spiculation represented by spike-like structures in the nodule’s boundary. In this paper, we develop a deep learning model based on a VGG16 architecture to locate the presence of spiculation in lung nodules from Computed Tomography images. In order to increase the expert’s confidence in the model output, we apply our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping attribution method to visualize areas of the image (spicules). Therefore, the attribution method is applied to the layer of the model that is responsible for the detection of the spiculation features. We show that the first layers of the network are specialized in detecting low-level features such as edges, the last convolutional layer detects the general area occupied by the nodule, and finally, we identify that spiculation structures are detected at an intermediate layer. We use three different metrics to support our findings.

KW - Artificial intelligence

KW - CAD

KW - Computer-aided detection

KW - Imaging informatics

KW - XAI

UR - http://www.scopus.com/inward/record.url?scp=85151000495&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85151000495&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-25082-8_30

DO - 10.1007/978-3-031-25082-8_30

M3 - Conference contribution

AN - SCOPUS:85151000495

SN - 9783031250811

T3 - Lecture Notes in Computer Science

SP - 457

EP - 471

BT - Computer Vision – ECCV 2022 Workshops, Proceedings

A2 - Karlinsky, Leonid

A2 - Michaeli, Tomer

A2 - Nishino, Ko

PB - Springer Science and Business Media Deutschland GmbH

T2 - 17th European Conference on Computer Vision, ECCV 2022

Y2 - 23 October 2022 through 27 October 2022

ER -