We consider a multihop sensor network in which sensors sample physical processes of interest and transmit the sample values in packets to a single destination. The destination remotely estimates the physical processes at the sensors from the received packets. The performance of remote estimation is measured in terms of the Age of Incorrect Information (AoII), which is a recently proposed goal-oriented communication metric. Our problem is to design multihop scheduling policies that minimize the average AoII (AAoII) across sensors under interference constraints. In order to characterize the minimum, we first consider centralized scheduling policies with knowledge of instantaneous AoII of the sensors. We formulate the multihop scheduling problem as a restless multiarmed bandit problem (RMAB) under the restriction that only a single packet from a sensor is allowed in the network at a time and under a complete interference model. We show that the optimal policy for a single-sensor problem obtained from the RMAB has a threshold structure and analytically characterize its AAoII. We also show that the RMAB is Whittle indexable and derive the indices in closed form. The Whittle indices are used to obtain an index policy for benchmarking multihop scheduling policies. A heuristic policy based on the index policy, which uses a belief of AAoII is then proposed. We compare the AAoII of the index policy, heuristic as well as Age-Difference policy from prior work using simulations. We also explore extensions to 𝐾-hop interference models, lossy link scenarios, and packet preemption cases. The proposed policies are shown to have better AAoII performance compared with Age-Difference policies.