Child Abuse & Neglect 109 (2020) 104755 Available online 16 October 2020 0145-2134/© 2020 Elsevier Ltd. All rights reserved. Can artifcial intelligence achieve human-level performance? A pilot study of childhood sexual abuse detection in self-fgure drawings Limor Kissos a , Limor Goldner a , Moshe Butman b , Niv Eliyahu b , Rachel Lev-Wiesel a, b, * a Emili Sagol Creative Arts Therapies Research Center, University of Haifa, Israel b ANIMA-EY LTD, Rishon Lezion, Israel A R T I C L E INFO Keywords: Child sexual abuse Drawing assessment Convolutional neural networks Artifcial intelligence ABSTRACT Childhood sexual abuse (CSA) is a worldwide phenomenon that has negative long-term conse- quences for the victims and their families, and inficts a considerable economic toll on society. One of the main diffculties in treating CSA is victims’ reluctance to disclose their abuse, and the failure of professionals to detect it when there is no forensic evidence (Bottoms et al., 2014; McElvaney, 2013). Estimated disclosure rates for child sexual abuse based on retrospective adult reports range from 23 % to 45 % (e.g., Bottoms et al., 2014). This study reports the four stages in the development of a Convolutional Neural Network (CNN) system designed to detect abuse in self-fgure drawings: (1) A preliminary study to build a Gender CNN; (2) Expert-level performance evaluation, (3) validation of the CSA CNN, (4) testing of the CSA CNN model. The fndings indicate that the Gender CNN achieved 88 % detection accuracy and outperformed the CSA CNN by 19 percentage points. The CSA CNN achieved 72 % accuracy on the test set with 80 % pre- cision and 79 % recall for the abuse class prediction. However, human experts outperformed the CSA CNN by 16 percentage points, probably due to the complexity of the task. These preliminary results suggest that CNN, when further developed, can contribute to the detection of child sexual abuse. 1. Introduction Sexual abuse is reported worldwide and has long-term negative physical and psychological consequences including behavioral and social problems (Peltzer and Pengpid, 2016), diffculties in emotional regulation (Chang, Kaczkurkin, McLean, & Foa, 2018), distress (anxiety disorders and depression), dissociative and post-traumatic disorders (H´ ebert, Langevin, & Daigneault, 2016), and chronic diseases such as fbromyalgia and Crohn’s disease (Gerber, Bogdan, Haskell, & Scioli, 2018; Steine et al., 2017). Despite the fact that early sexual abuse detection is critical (Daigneault, V´ ezina-Gagnon, Bourgeois, Esposito, & H´ ebert, 2017; Morais, Alexander, Fix, & Burkhart, 2018), CSA victims’ reluctance to disclose appear to be an international phenomenon (Berry & Rutledge, 2016; Bottoms et al., 2014; McElvaney, Greene, & Hogan, 2014; Collin-V´ ezina, De La Sablonni` ere-Griffn, Palmer, & Milne, 2015; Hu et al., 2018). Furthermore, professionals may fail to identify abuse when there are no visible signs (Pelisoli, Herman, & Dell’Aglio, 2015; Powell & * Corresponding author. E-mail address: rlev@univ.haifa.ac.il (R. Lev-Wiesel). Contents lists available at ScienceDirect Child Abuse & Neglect journal homepage: www.elsevier.com/locate/chiabuneg https://doi.org/10.1016/j.chiabu.2020.104755 Received 20 March 2019; Received in revised form 5 August 2020; Accepted 17 September 2020