Toxicological Research

Korean Society of Toxicology & Korea Environmental Mutagen Society (한국독성학회)
권호 표지
DOI QR코드

DOI QR Code

A comparative study on the implementation of deep learning algorithms for detection of hepatic necrosis in toxicity studies

  • Ji‑Hee Hwang (Toxicologic Pathology Research Group, Department of Advanced Toxicology Research, Korea Institute of Toxicology) ;
  • Minyoung Lim (Toxicologic Pathology Research Group, Department of Advanced Toxicology Research, Korea Institute of Toxicology) ;
  • Gyeongjin Han (Toxicologic Pathology Research Group, Department of Advanced Toxicology Research, Korea Institute of Toxicology) ;
  • Heejin Park (Toxicologic Pathology Research Group, Department of Advanced Toxicology Research, Korea Institute of Toxicology) ;
  • Yong‑Bum Kim (Department of Advanced Toxicology Research, Korea Institute of Toxicology) ;
  • Jinseok Park (Research & Development Team, LAC Inc) ;
  • Sang‑Yeop Jun (Research & Development Team, LAC Inc) ;
  • Jaeku Lee (Research & Development Team, LAC Inc) ;
  • Jae‑Woo Cho (Toxicologic Pathology Research Group, Department of Advanced Toxicology Research, Korea Institute of Toxicology)
  • Received : 2022.12.18
  • Accepted : 2023.02.20
  • Published : 2023.07.15
⟨ Previous Next ⟩

Abstract

Deep learning has recently become one of the most popular methods of image analysis. In non-clinical studies, several tissue slides are generated to investigate the toxicity of a test compound. These are converted into digital image data using a slide scanner, which is then studied by researchers to investigate abnormalities, and the deep learning method has been started to adopt in this study. However, comparative studies evaluating different deep learning algorithms for analyzing abnormal lesions are scarce. In this study, we applied three algorithms, SSD, Mask R-CNN, and DeepLabV3+, to detect hepatic necrosis in slide images and determine the best deep learning algorithm for analyzing abnormal lesions. We trained each algorithm on 5750 images and 5835 annotations of hepatic necrosis including validation and test, augmented with 500 image tiles of 448×448 pixels. Precision, recall, and accuracy were calculated for each algorithm based on the prediction results of 60 test images of 2688×2688 pixels. The two segmentation algorithms, DeepLabV3+ and Mask R-CNN, showed over 90% of accuracy (0.94 and 0.92, respectively), whereas SSD, an object detection algorithm, showed lower accuracy. The trained DeepLabV3+ outperformed all others in recall while also successfully separating hepatic necrosis from other features in the test images. It is important to localize and separate the abnormal lesion of interest from other features to investigate it on a slide level. Therefore, we suggest that segmentation algorithms are more appropriate than object detection algorithms for use in the pathological analysis of images in non-clinical studies.

Keywords

Acknowledgement

We thank Ga-Hyun Kim and Ji-soo Yang for annotating the hepatic fbrosis on all the image data.

References

  1. Pantanowitz L, Sinard JH, Henricks WH, Fatheree LA, Carter AB, Contis L et al (2013) Validating whole slide imaging for diagnostic purposes in pathology: guideline from the College of American Pathologists Pathology and Laboratory Quality Center. Arch Pathol Lab Med 137:1710-1722. https://doi.org/10.5858/arpa.2013-0093-CP
  2. Zarella MD, Bowman D, Aefner F, Farahani N, Xthona A, Absar SF et al (2019) A practical guide to whole slide imaging: a white paper from the digital pathology association. Arch Pathol Lab Med 143:222-234. https://doi.org/10.5858/arpa.2018-0343-RA
  3. Kuklyte J, Fitzgerald J, Nelissen S, Wei H, Whelan A, Power A et al (2021) Evaluation of the use of single- and multi-magnification convolutional neural networks for the determination and quantitation of lesions in nonclinical pathology studies. Toxicol Pathol 49:815-842. https://doi.org/10.1177/0192623320986423
  4. Center for Food Safety and Applied Nutrition, Office of Food Additive Safety. (2003) Subchronic toxicity studies with rodents. Redbook. 2000. U.S. Food and Drug Administration. IV:C.4.a
  5. Mehrvar S, Himmel LE, Babburi P, Goldberg AL, Gufroy M, Janardhan K et al (2021) Deep learning approaches and applications in toxicologic histopathology: current status and future perspectives. J Pathol Inform 1:12-42. https://doi.org/10.4103/jpi.jpi_36_21
  6. Redmon J, Divvala S, Girshick R, Farhadi A (2016) You only look once: Unified, real-time object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp 779-788. https://doi.org/10.1109/CVPR.2016.91
  7. Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu C-Y et al (2016) SSD: single shot MultiBox detector. In: Leibe B, Matas J, Sebe N, Welling M (eds) Computer vision - ECCV 2016. ECCV 2016. Lecture Notes in Computer Science, vol 9905. Springer, Cham. pp 21-37. https://doi.org/10.48550/arXiv.1512.02325
  8. Lin TY, Goyal P, Girshick R, He K, Dollar P (2017) Focal loss for dense object detection. In: Proceedings of the IEEE international conference on computer vision. pp 2980-2988. https://doi.org/10.48550/arXiv.1708.02002
  9. Bukowy JD, Dayton A, Cloutier D, Manis AD, Staruschenko A, Lombard JH et al (2018) Region-based convolutional neural nets for localization of glomeruli in trichrome-stained whole kidney sections. J Am Soc Nephrol 29:2081-2088. https://doi.org/10.1681/ASN.2017111210
  10. Carboni E, Marxfeld H, Tuoken H, Klukas C, Eggers T, Groters S et al (2021) A workflow for the performance of the differential ovarian follicle count using deep neuronal networks. Toxicol Pathol 49:843-850. https://doi.org/10.1177/0192623320969130
  11. Hu F, Schutt L, Kozlowski C, Regan K, Dybdal N, Schutten MM (2020) Ovarian toxicity assessment in histopathological images using deep learning. Toxicol Pathol 48:350-361. https://doi.org/10.1177/0192623319877871
  12. Long J, Shelhamer E, Darrell T (2017) Fully convolutional networks for semantic segmentation. IEEE Trans Pattern Anal Mach Intell 39:640-651. https://doi.org/10.48550/arXiv.1411.4038
  13. Ronneberger O, Fischer P (2015) Brox TUnet: Convolutional Networks for Biomedical Image Segmentation. In: International Conference on Medical image computing and computer-assisted intervention. pp 234-241. https://doi.org/10.48550/arXiv.1505.04597
  14. Chen L, Zhu Y, Papandreou G, Schroff F, Adam H (2018) Encoder-decoder with atrous separable convolution for semantic image segmentation. Computer Vision-ECCV 2018:833-851. https://doi.org/10.1007/978-3-030-01234-2_49
  15. He K, Gkioxari G, Dollar P, Girshick R (2017) Mask R-CNN. In: Proceedings of the IEEE international conference on computer vision (ICCV). pp 2980-2988. https://doi.org/10.1109/iccv.2017.322
  16. Hwang JH, Kim HJ, Park H, Lee BS, Son HY, Kim YB et al (2022) Implementation and practice of deep learning-based instance segmentation algorithm for quantification of hepatic fibrosis at whole slide level in Sprague-Dawley rats. Toxicol Pathol 50:186-196. https://doi.org/10.1177/01926233211057128
  17. Baek EB, Hwang JH, Park H, Lee BS, Son HY, Kim YB et al (2022) Artificial intelligence-assisted image analysis of acetaminophen-induced acute hepatic injury in Sprague-Dawley rats. Diagnostics 12:1478. https://doi.org/10.3390/diagnostics12061478
  18. Grattagliano I, Bonfrate L, Diogo CV, Wang HH, Wang DQ, Portincasa P (2009) Biochemical mechanisms in drug-induced liver injury: certainties and doubts. World J Gastroenterol 15:4865-4876. https://doi.org/10.3748/wjg.15.4865
  19. Ramachandran R, Kakar S (2009) Histological patterns in drug-induced liver disease. J Clin Pathol 62:481-492. https://doi.org/10.1136/jcp.2008.058248
  20. Black M (1980) Acetaminophen hepatotoxicity. Gastroenterology 78:382-392. https://doi.org/10.1016/0016-5085(80)90593-4
  21. Papackova Z, Heczkova M, Dankova H, Sticova E, Lodererova A, Bartonova L et al (2018) Silymarin prevents acetaminophen-induced hepatotoxicity in mice. PLoS ONE 13:e0191353. https://doi.org/10.1371/journal.pone.0191353
  22. Prepared by the Animal Facilities Standards Committee of the Animal Care Panel (2022) Guide for laboratory animal facilities and care. ILAR J. https://doi.org/10.1093/ilar/ilac012
  23. MFDS (2017) Guidelines for toxicity tests of pharmaceuticals. No. 2017-71. Ministry of Food and Drug Safety, Sejong, Republic of Korea
  24. Abdel-Zaher AO, Abdel-Hady RH, Mahmoud MM, Farrag MM (2008) The potential protective role of alpha-lipoic acid against acetaminophen-induced hepatic and renal damage. Toxicology 243:261-270. https://doi.org/10.1016/j.tox.2007.10.010
  25. Yousef MI, Omar SA, El-Guendi MI, Abdelmegid LA (2010) Potential protective effects of quercetin and curcumin on paracetamol-induced histological changes, oxidative stress, impaired liver and kidney functions and haematotoxicity in rat. Food Chem Toxicol 48:3246-3261. https://doi.org/10.1016/j.fct.2010.08.034
  26. TorchVision maintainers and contributors (2016) TorchVision: PyTorch's Computer Vision library [Computersoftware]. GitHub repository, https://github.com/pytorch/vision. Accessed 14 Aug 2022
  27. Zhang J (2019) pytorch-deeplab-xception package [Computer software]. GitHub repository, https://github.com/jfzhang95/pytor ch-deeplab-xception. Accessed 14 Aug 2022
  28. DeGroot M (2019) amdegroot SSD.pytorch package [Computer software]. GitHub repository, https://github.com/amdegroot/ssd. pytorch. Accessed 14 Aug 2022
  29. Goutte C, Gaussier E (2005) A probabilistic interpretation of precision, recall and F-score, with implication for evaluation. In: Losada DE, Fernandez-Luna JM (eds) Advances in information retrieval. Lect notes Comput Sci. ECIR 2005, vol 3408. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-31865-1_25
  30. Aubreville M, Bertram CA, Marzahl C, Gurtner C, Dettwiler M, Schmidt A et al (2020) Deep learning algorithms out-perform veterinary pathologists in detecting the mitotically most active tumor region. Sci Rep 10:16447. https://doi.org/10.1038/s41598-020-73246-2
  31. Horai Y, Kakimoto T, Takemoto K, Tanaka M (2017) Quantitative analysis of histopathological findings using image processing software. J Toxicol Pathol 30:351-358. https://doi.org/10.1293/tox.2017-0031
  32. Horai Y, Mizukawa M, Nishina H, Nishikawa S, Ono Y, Takemoto K et al (2019) Quantifcation of histopathological findings using a novel image analysis platform. J Toxicol Pathol 32:319-327. https://doi.org/10.1293/tox.2019-0022
  33. Pischon H, Mason D, Lawrenz B, Blanck O, Frisk AL, Schorsch F et al (2021) Artifcial intelligence in toxicologic pathology: quantitative evaluation of compound-induced hepatocellular hypertrophy in rats. Toxicol Pathol 49:928-937. https://doi.org/10.1177/0192623320983244