A neural cell automated analysis system based on pathological specimens in a gerbil brain ischemia model

Amid rising health awareness, natural products which has milder effects than medical drugs are becoming popular. However, only few systems can quantitatively assess their impact on living organisms. Therefore, we developed a deep-learning system to automate the counting of cells in a gerbil model, aiming to assess a natural product’s effectiveness against ischemia.

Methods:

The image acquired from paraffin blocks containing gerbil brains was analyzed by a deep-learning model (fine-tuned Detectron2).

Results:

The counting system achieved a 79%-positive predictive value and 85%-sensitivity when visual judgment by an expert was used as ground truth.

Conclusions:

Our system evaluated hydrogen water’s potential against ischemia and found it potentially useful, which is consistent with expert assessment. Due to natural product’s milder effects, large data sets are needed for evaluation, making manual measurement labor-intensive. Hence, our system offers a promising new approach for evaluating natural products.

Key words
Deep Learning; Artificial Intelligence; Ischemia; Cerebral Cortex; Pathology

Authorship

Eri Katsumata

Conception and design

Acquisition and analysis of data

Manuscript writing

MSs

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

https://orcid.org/0009-0001-9015-4083

Abhishek Kumar Ranjan

Acquisition and analysis of data

Technical procedure

MSs

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

https://orcid.org/0009-0000-5925-2526

Yoshihiko Tashima

Substantive scientific and intellectual contributions to the study

Critical revision

PhDs

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

https://orcid.org/0009-0002-1753-6293

Takayuki Takahata

Substantive scientific and intellectual contributions to the study

Critical revision

MSs

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

https://orcid.org/0009-0001-6214-9500

Toshiyuki Sato

Substantive scientific and intellectual contributions to the study

Critical revision

MSs

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

https://orcid.org/0000-0003-0127-3521

Motoaki Kobayashi

Substantive scientific and intellectual contributions to the study

Critical revision

Sept.Sapie Co. – Tokyo, Japan.Sept.Sapie Co.JapanTokyo, JapanSept.Sapie Co. – Tokyo, Japan.

https://orcid.org/0009-0009-1200-9348

Masami Ishii

Substantive scientific and intellectual contributions to the study

Critical revision

Sept.Sapie Co. – Tokyo, Japan.Sept.Sapie Co.JapanTokyo, JapanSept.Sapie Co. – Tokyo, Japan.

https://orcid.org/0009-0004-5574-5170

Toyomi Takahashi

Substantive scientific and intellectual contributions to the study

Critical revision

PhDs

Sept.Sapie Co. – Tokyo, Japan.Sept.Sapie Co.JapanTokyo, JapanSept.Sapie Co. – Tokyo, Japan.

https://orcid.org/0009-0009-8309-325X

Asahi Oda

Substantive scientific and intellectual contributions to the study

Acquisition and analysis of data

Technical procedure

MSs

Nippon Veterinary and Life Science University – School of Veterinary Nursing and Technology – Department of Basic Science – Tokyo, Japan.Nippon Veterinary and Life Science UniversityJapanTokyo, JapanNippon Veterinary and Life Science University – School of Veterinary Nursing and Technology – Department of Basic Science – Tokyo, Japan.

https://orcid.org/0009-0001-2083-7372

Momoko Hirano

Substantive scientific and intellectual contributions to the study

Technical procedure

MSs

https://orcid.org/0009-0007-4347-9372

Yoji Hakamata

Conception and design

Substantive scientific and intellectual contributions to the study

Acquisition and analysis of data

Technical procedure

PhD

https://orcid.org/0000-0002-0138-6692

Kazuhisa Sugai

Substantive scientific and intellectual contributions to the study

Critical revision

PhDs

https://orcid.org/0000-0003-2152-5394

Eiji Kobayashi ^** Autora correspondente: eijikoba@jikei.ac.jp

Conception and design

Substantive scientific and intellectual contributions to the study

Manuscript writing

Final approval of the version to be published

MD, PhD

Kobayashi Regenerative Research Institute – Wakayama, Japan.Kobayashi Regenerative Research InstituteJapanWakayama, JapanKobayashi Regenerative Research Institute – Wakayama, Japan.

https://orcid.org/0000-0001-8617-3778

* Autora correspondente: eijikoba@jikei.ac.jp

Section editor: Eberval Gadelha https://orcid.org/0000-0001-9094-6622

Conflict of interest

Nothing to declare.

SCIMAGO INSTITUTIONS RANKINGS

Sysmex Co. – Hyogo, Japan.Sysmex CoJapanHyogo, JapanSysmex Co. – Hyogo, Japan.

Sept.Sapie Co. – Tokyo, Japan.Sept.Sapie Co.JapanTokyo, JapanSept.Sapie Co. – Tokyo, Japan.

Kobayashi Regenerative Research Institute – Wakayama, Japan.Kobayashi Regenerative Research InstituteJapanWakayama, JapanKobayashi Regenerative Research Institute – Wakayama, Japan.

Figures | Formulas

Figures (10)
Formulas (1)

Thumbnail

Figure 1
Specimen used in the study.

Thumbnail

Figure 2
Model’s input information. (a) Sample annotations by an expert. (b) Sample annotation. (c) Annotation intuition.

Thumbnail

Figure 3
The output of data augmentation. (a) Original image. (b) Horizontal flip. (c) Vertical flip. (d) Positive brightness. (e) Negative brightness. (f) Positive Contrast. (g) Negative Contrast. (h) Sharpness.

Thumbnail

Figure 4
Threshold selection. Fix the value of one indicator and determine the appropriate threshold for the other indicator. (a) When confidence score is fixed. (b) When Intersection over Union (IoU) is fixed.

Thumbnail

Figure 5
Positive predictive value, sensitivity, and F1-score in five different models. The results of the artificial intelligence system represent the averages of five trials, with error bars indicating one standard deviation.

Thumbnail

Figure 6
Comparison of cell counts by experts, novices, and the artificial intelligence (AI) system. Presented are the averages of five trials with error bars representing one standard deviation. (a) The cell counts. C1, C2, C3, C4, and C8 denote individual identifiers, while L and R indicate the left hemisphere and right hemisphere, respectively. (b) The left-right ratios obtained by dividing the cell counts in specific regions of the left hemisphere by those in the corresponding regions of the right hemisphere.

Thumbnail

Figure 7
Relationship between the number of counted cells and the time taken for counting.

Thumbnail

Figure 8
Left-right ratios in each group. The artificial intelligence (AI) results represent the averages of five trials, with error bars indicating one standard deviation. (a) Control group. (b) Group induced with ischemic neuronal cell death. (c) Group induced with ischemic neuronal cell death and administered hydrogen water.

Thumbnail

Figure 9
Regions detected as false positives. Red frames represent regions identified as cells by an expert, while black frames indicate regions identified as cells by the artificial intelligence system. The numerical values in black frames denote the confidence score for each region.

Thumbnail

Figure 10
Variability in cell counts by the artificial intelligence system. Regions enclosed by squares were identified as cells. (a) Annotation results based on visual inspection. (b-d) Results of cell counts by the artificial intelligence system from the first to the third trial.

Thumbnail

(1)

Figure 1 Specimen used in the study.

Source: Elaborated by the authors.

Figure 2 Model’s input information. (a) Sample annotations by an expert. (b) Sample annotation. (c) Annotation intuition.

Source: Elaborated by the authors.

Figure 3 The output of data augmentation. (a) Original image. (b) Horizontal flip. (c) Vertical flip. (d) Positive brightness. (e) Negative brightness. (f) Positive Contrast. (g) Negative Contrast. (h) Sharpness.

Source: Elaborated by the authors.

Figure 4 Threshold selection. Fix the value of one indicator and determine the appropriate threshold for the other indicator. (a) When confidence score is fixed. (b) When Intersection over Union (IoU) is fixed.

Source: Elaborated by the authors.

Figure 5 Positive predictive value, sensitivity, and F1-score in five different models. The results of the artificial intelligence system represent the averages of five trials, with error bars indicating one standard deviation.

Source: Elaborated by the authors.

Figure 6 Comparison of cell counts by experts, novices, and the artificial intelligence (AI) system. Presented are the averages of five trials with error bars representing one standard deviation. (a) The cell counts. C1, C2, C3, C4, and C8 denote individual identifiers, while L and R indicate the left hemisphere and right hemisphere, respectively. (b) The left-right ratios obtained by dividing the cell counts in specific regions of the left hemisphere by those in the corresponding regions of the right hemisphere.

Source: Elaborated by the authors.

Figure 7 Relationship between the number of counted cells and the time taken for counting.

AI: Artificial intelligence. Source: Elaborated by the authors.

Figure 8 Left-right ratios in each group. The artificial intelligence (AI) results represent the averages of five trials, with error bars indicating one standard deviation. (a) Control group. (b) Group induced with ischemic neuronal cell death. (c) Group induced with ischemic neuronal cell death and administered hydrogen water.

Source: Elaborated by the authors.

Figure 9 Regions detected as false positives. Red frames represent regions identified as cells by an expert, while black frames indicate regions identified as cells by the artificial intelligence system. The numerical values in black frames denote the confidence score for each region.

Source: Elaborated by the authors.

Figure 10 Variability in cell counts by the artificial intelligence system. Regions enclosed by squares were identified as cells. (a) Annotation results based on visual inspection. (b-d) Results of cell counts by the artificial intelligence system from the first to the third trial.

Source: Elaborated by the authors.

(1)

I o U = (Area of Overlap) / (Area of Union)

How to cite

copy

Sociedade Brasileira para o Desenvolvimento da Pesquisa em Cirurgia https://actacirbras.com.br/ - São Paulo - SP - Brazil
E-mail: actacirbras@gmail.com

Acompanhe os números deste periódico no seu leitor de RSS

Versão para download de PDF

PDF

Inglês

Versões e tradução automática

Versão original do texto

English

Tradução automática

Google Translator
Microsoft Translator

Como citar

RIS

BIBTEX

Outros formatos de citação e exportação:

SciELO - Scientific Electronic Library Online
Rua Dr. Diogo de Faria, 1087 – 9º andar – Vila Clementino 04037-003 São Paulo/SP - Brasil
E-mail: scielo@scielo.org

Leia a Declaração de Acesso Aberto

Métricas

A neural cell automated analysis system based on pathological specimens in a gerbil brain ischemia model

PlumX

Mensagem

[1] * Autora correspondente: eijikoba@jikei.ac.jp