This work presents the characterization results and methods used to determine the reconstruction of a 22Na point source placed in a coincidence setup. For this purpose, a novel, fully-integrated, reconfigurable feed-forward artificial neural network was implemented in a 16-nm FinFET CMOS technology node. The artificial neural network uses timestamps obtained from 10 on-chip time-to-digital converters. The network topology can be reconfigured at any time within design limits. After appropriate training, the artificial neural network successfully distinguished six different radioactive source positions placed between the two photodetectors along the X axis that fed the time-to-digital converters. Thanks to its ability to filter frames without any coincidence and to increase the overall sensitivity, the on-chip approach significantly reduces both system complexity and data throughput.