A deep neural network to assess spontaneous pain from mouse facial expressions

Grimace scales quantify characteristic facial expressions associated with spontaneous pain in rodents and other mammals. However, these scales have not been widely adopted largely because of the time and effort required for highly trained humans to manually score the images. Convoluted neural networks were recently developed that distinguish individual humans and objects in images. Here, we trained one of these networks, the InceptionV3 convolutional neural net, with a large set of human-scored mouse images. Output consists of a binary pain/no-pain assessment and a confidence score. Our automated Mouse Grimace Scale integrates these two outputs and is highly accurate (94%) at assessing the presence of pain in mice across different experimental assays. In addition, we used a novel set of “pain” and “no pain” images to show that automated Mouse Grimace Scale scores are highly correlated with human scores (Pearson’s r = 0.75). Moreover, the automated Mouse Grimace Scale classified a greater proportion of images as “pain” following laparotomy surgery when compared to animals receiving a sham surgery or a post-surgical analgesic. Together, these findings suggest that the automated Mouse Grimace Scale can eliminate the need for tedious human scoring of images and provide an objective and rapid way to quantify spontaneous pain and pain relief in mice.

[1]  E. Troncy,et al.  Assessing experimental visceral pain in dairy cattle: A pilot, prospective, blinded, randomized, and controlled study focusing on spinal pain proteomics. , 2014, Journal of dairy science.

[2]  E. Holden,et al.  Evaluation of facial expression in acute pain in cats. , 2014, The Journal of small animal practice.

[3]  S. Thrun,et al.  Corrigendum: Dermatologist-level classification of skin cancer with deep neural networks , 2017, Nature.

[4]  Fan Xia,et al.  Rapid and accurate intraoperative pathological diagnosis by artificial intelligence with deep learning technology. , 2017, Medical hypotheses.

[5]  J. Mogil,et al.  What should we be measuring in behavioral studies of chronic pain in animals? , 2004, Pain.

[6]  C. Lindegaard,et al.  An equine pain face , 2014, Veterinary anaesthesia and analgesia.

[7]  Amy L. Miller,et al.  Efficacy of Intrathecal Morphine in a Model of Surgical Pain in Rats , 2016, PloS one.

[8]  Ognjen Rudovic,et al.  Personalized Automatic Estimation of Self-Reported Pain Intensity from Facial Expressions , 2017, 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).

[9]  J. Mogil,et al.  Using the Mouse Grimace Scale to reevaluate the efficacy of postoperative analgesics in laboratory mice. , 2012, Journal of the American Association for Laboratory Animal Science : JAALAS.

[10]  P. Carrive,et al.  Peripheral and Central Neuroinflammatory Changes and Pain Behaviors in an Animal Model of Multiple Sclerosis , 2016, Front. Immunol..

[11]  Hui Li,et al.  Deep learning in breast cancer risk assessment: evaluation of convolutional neural networks on a clinical dataset of full-field digital mammograms , 2017, Journal of medical imaging.

[12]  M. Minero,et al.  Development of the Horse Grimace Scale (HGS) as a Pain Assessment Tool in Horses Undergoing Routine Castration , 2014, PloS one.

[13]  Amy L. Miller,et al.  The Influence of Isoflurane Anaesthesia on the Rat Grimace Scale , 2016, PloS one.

[14]  Kenneth D. Craig,et al.  Pain expression in neonates: facial action and cry , 1987, Pain.

[15]  Thomas Frauenfelder,et al.  Deep Learning in Mammography: Diagnostic Accuracy of a Multipurpose Image Analysis Software in the Detection of Breast Cancer , 2017, Investigative radiology.

[16]  Amy L. Miller,et al.  The Assessment of Post-Vasectomy Pain in Mice Using Behaviour and the Mouse Grimace Scale , 2012, PloS one.

[17]  Loren J. Martin,et al.  The Rat Grimace Scale: A partially automated method for quantifying pain in the laboratory rat via facial expressions , 2011, Molecular pain.

[18]  Ryan P. Adams,et al.  Mapping Sub-Second Structure in Mouse Behavior , 2015, Neuron.

[19]  Peter Robinson,et al.  Estimating Sheep Pain Level Using Facial Action Unit Detection , 2017, 2017 12th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2017).

[20]  L. Shientag.  Efficacy of Common Analgesics for Postsurgical Pain in Rats. , 2016, Journal of the American Association for Laboratory Animal Science : JAALAS.

[21]  K. Craig,et al.  Coding of facial expressions of pain in the laboratory mouse , 2010, Nature Methods.

[22]  Yuanjie Zheng,et al.  Breast Cancer Multi-classification from Histopathological Images with Structured Deep Learning Model , 2017, Scientific Reports.

[23]  Jurgen E Schneider,et al.  Refinement of analgesia following thoracotomy and experimental myocardial infarction using the Mouse Grimace Scale , 2015, Experimental physiology.

[24]  Natalia Antropova,et al.  A deep feature fusion methodology for breast cancer diagnosis demonstrated on three imaging modality datasets , 2017, Medical physics.

[25]  Sergey Ioffe,et al.  Rethinking the Inception Architecture for Computer Vision , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[26]  David Borsook,et al.  Lost but making progress—Where will new analgesic drugs come from? , 2014, Science Translational Medicine.

[27]  Hayit Greenspan,et al.  Deep learning with non-medical training used for chest pathology identification , 2015, Medical Imaging.

[28]  Mustafa Eral,et al.  Assessment of pain in mouse facial images , 2016, 2016 20th National Biomedical Engineering Meeting (BIYOMUT).

[29]  Amy L. Miller,et al.  The Mouse Grimace Scale: A Clinically Useful Tool? , 2015, PloS one.

[30]  P. Flecknell,et al.  Evaluation of EMLA Cream for Preventing Pain during Tattooing of Rabbits: Changes in Physiological, Behavioural and Facial Expression Responses , 2012, PloS one.

[31]  J. Mogil.  Animal models of pain: progress and challenges , 2009, Nature Reviews Neuroscience.

[32]  A. Faden,et al.  Cell cycle inhibition limits development and maintenance of neuropathic pain following spinal cord injury , 2016, Pain.

[33]  B. Seifert,et al.  Improving animal welfare using continuous nalbuphine infusion in a long-term rat model of sepsis , 2017, Intensive Care Medicine Experimental.

[34]  Richard Grossman,et al.  The animals , 1983 .

[35]  Katherine B. Martin,et al.  Facial Action Coding System , 2015 .

[36]  Michael Blum,et al.  High-Throughput Classification of Radiographs Using Deep Convolutional Neural Networks , 2016, Journal of Digital Imaging.

[37]  M. Leach,et al.  Development of a Piglet Grimace Scale to Evaluate Piglet Pain Using Facial Expressions Following Castration and Tail Docking: A Pilot Study , 2017, Front. Vet. Sci..

[38]  D. Pang,et al.  Real-time application of the Rat Grimace Scale as a welfare refinement in laboratory rats , 2016, Scientific Reports.

[39]  Andrew Shea,et al.  DeepFaceLIFT: Interpretable Personalized Models for Automatic Estimation of Self-Reported Pain , 2017, AffComp@IJCAI.

[40]  Tracy L. Niedzielko,et al.  Application of the Rat Grimace Scale as a Marker of Supraspinal Pain Sensation after Cervical Spinal Cord Injury. , 2017, Journal of neurotrauma.

[41]  Andrew S.C. Rice,et al.  Animal models and the prediction of efficacy in clinical trials of analgesic drugs: A critical appraisal and call for uniform reporting standards , 2008, PAIN.

[42]  K. Stafford,et al.  Coding and quantification of a facial expression for pain in lambs , 2016, Behavioural Processes.

[43]  Sebastian Thrun,et al.  Dermatologist-level classification of skin cancer with deep neural networks , 2017, Nature.