This figure shows some positive (rows 1-5) and negative (rows 6 and 7) highlights of image conversion, chosen as exemplary for various qualitative observations made. Column 1 (left) shows ground truth simulated computed tomography (CT) images. Column 2 shows simulated B-mode images. Columns 3-8 show the output of the ML model when trained on different stages of the ultrasound signal processing pipeline: after input signal removal, time gain compensation, frequency filtering, envelope detection, log compression, and upsampling (results using raw scan lines as input are omitted due to their inferior performance, see Figure 2). The final column shows generated CT images in red (using the average image intensities of the 6 previous columns, normalized to cover the entire dynamic range) superimposed on a B-Mode image.