Fusarium head blight (FHB) is one of the most significant diseases affecting wheat worldwide. An important factor complicating the management of FHB and the development of wheat varieties with resistance to FHB is that disease phenotyping is time consuming, laborious and resource intensive. The problems considered in this work are the automated detection of FHB disease symptoms expressed on a wheat plant, and estimation of the severity of the disease, which is determined by the ratio of the total number of infected spikelets to the total number of spikelets on the same wheat head. The data used to generate the results are 3-dimensional (3D) multispectral point clouds (PC), which are 3D collections of points representing objects in a 3D space, where each point is associated with a red, green, blue (RGB), and near-infrared (NIR) measurement. The PCs were used to develop convolutional neural networks (CNN), which are models that excel in mapping complex input data to specific class labels. Using a multispectral 3D scanner, two datasets were created by acquiring 336 and 200 wheat plant PCs, respectively, each consisting of both healthy and FHB-diseased samples at various infection stages (i.e., days post inoculation). The first dataset was used to develop novel and efficient 3D CNNs for FHB detection, and the second dataset was used to develop unique and reliable 3D CNNs to estimate both the total number of spikelets and the total number of infected spikelets, thereby paving the way for an automated approach to severity estimation. Moreover, the influence of the multispectral information on performance was evaluated, and our results showed the dominance of the RGB channels over both the NIR and the RGB and NIR channels combined. Future use of this automated system will help our team to accelerate the research improving FHB management practices and detection, thereby characterizing, and employing genetic resistance to help build better and elite wheat cultivars.