*2.1. Participants*

Patient PL518 and four other patients participated in this specific study. All were recruited as part of a larger study of PCA stroke (the Back of the Brain (BoB) project, described in [40]). 46 controls were

included in the BoB project. All participants provided written, informed consent, and the project was approved by the ethical committees of Manchester (North West Research Ethics Committee; MREC 01/8/094) and UCL (London Queen Square Research Ethics Committee, UCL; 16/EM/0348). See Table 1 for demographics and background data for the included patients and controls. Additional background data as well as raw scores on the perceptual and neuropsychological tests can be found in Table S1.

**Table 1.** Demographic and lesion information for the five included patients and controls, and scores on basic tests. Handedness was measured by the short form of the Edinburgh Handedness Inventory (EHI) [41]; depression was measured with the short version of the Geriatric Depression Scale (GDS-15 [42]). General cognition was screened with the Oxford Cognitive Screen (OCS) [43], and the number of impaired subtests are listed. Digit span forward and backward was measured with the WAIS-IV UK [44] and total scores are listed. Basic motor reaction time (RT) was measured by responding to a bar of light presented horizontally on a screen (test described in [40]).


PL518 su ffered a bilateral PCA stroke 35 months before the current investigation. At that time, he had corrected to normal visual acuity and a slight visual field defect primarily a ffecting the parafovea of the upper left quadrant (see Table S1 for acuity and visual field data for all participants). He reported problems with seeing colors following his stroke and scored outside the normal range on a formal test of color perception (Farnsworth D-15 [45]). His intermediate vision (assessed with subtests from the L-POST [46]) was largely uncompromised, except for di fficulties with figure-ground segmentation. His basic response time (RT) to visual stimuli (test described in [40]) was una ffected, and his auditory digit span forwards and backwards (WAIS-IV UK [44]) were within the normal range. PL518 reported severe problems in face recognition following his stroke and volunteered that he had problems recognizing his own face in the mirror. He also reported increased problems in finding his way around. Neuropsychological testing showed a clear deficit in face recognition a ffecting learning of new faces as well as judgment of familiarity and recognition of famous faces. In contrast, he performed within the normal range on several tests of object recognition, including perceptually challenging tests, with the notable exception of a memory test for houses (Cambridge House Memory Test, [47]) designed as an equivalent to the Cambridge Face Memory Test [48]. His word recognition accuracy was well within normal range, while response times in reading out loud and the e ffect of word length on RT was slightly but significantly elevated compared to controls.

In the context of the BoB project, PL518 spontaneously reported that his visual imagery was "gone" following his stroke, which led us to contact four additional patients either with similar lesions (bilateral PCA stroke) or a similar background (architect) for a follow-up interview about their visual imagery. The lesion location, size, and time since injury for these patients, as well as other background characteristics, are presented in Table 1 along with summary data from the control group.

### *2.2. Visual Imagery*

As PL 518 was the focus of the study, a long and in-depth interview was also carried out about his visual imagery before and after his stroke. In order to ge<sup>t</sup> more information about his ability to store visual information in his mind, he was asked to carry out the Rey–Osterrieth Complex Figure Test [49] at the end of the interview. The other patients did not complete this test.

All five patients were asked to complete a version of the Vividness of Visual Imagery Questionnaire (VVIQ-modified, [39]), that is a modified version of the VVIQ [50]. VVIQ-modified has 16 items where participants are to imagine various scenarios (e.g., a relative or friend, a rising sun) and rate the vividness of their visual image on a five-point Likert scale where 1 indicates no image at all and 5 indicates that the image is perfectly clear and vivid. Scores on the VVIQ-modified can range from 16 to 80, with 16 representing the lowest possible imagery score and 80 the highest possible imagery score. Various versions of the VVIQ have been validated, and the questionnaire has been shown to be a valid psychometric tool for measuring the vividness of visual imagery with both high construct validity and internal consistency reliability [51–53]. PL518 and PM024 performed the questionnaire in the lab while the other three patients were interviewed over the telephone. The four control patients were also asked four general questions about their visual imagery, to compare with the interview of PL518. They were asked to answer the following questions with yes, no, or don't know: (1) can you imagine things visually in your mind? (if no: do you sometimes experience brief flashes of imagery?); (2) would you say that your memories have a visual aspect to them in your mind?; (3) do you see visual images in dreams?; and (4) has your visual imagery changed following your stroke? The normal controls did not perform the visual imagery questionnaire.

### *2.3. Neuropsychological and Experimental Tests*

The BoB project is a comprehensive neuropsychological and imaging project investigating perceptual deficits following posterior brain injury [40]. A main aim of the overall project is to compare patient performance with faces, objects, and words. The main findings of the project are not ye<sup>t</sup> published (paper in preparation, [54]). Here we report data from the five included patients and controls on tests and experiments selected to be comparable across categories for faces, objects, and words, and these are briefly described below. These experiments, as well as all other tests included in the project, are described in full in [40]. The experimental tests were run on laptop computers with screen resolution of 1366 × 768, or on desktop computers with a screen resolution of 1920 × 1080.

### 2.3.1. Delayed Matching and Surprise Recognition of Words, Objects and Faces—The WOF Test

This novel paradigm is designed to test immediate and delayed memory for words, objects, and faces (WOF). In the first part (delayed matching), participants were asked to decide whether two sequentially presented images varying in size are the same or not. There were 48 trials for each stimulus type and both accuracy and RTs are measured. The second part (surprise recognition) followed after a short break (where participants performed an unrelated task, the Farnsworth D-15). Here, participants were asked to decide whether they saw the presented stimuli in the delayed matching task or not. There were 12 trials, and accuracy and RTs were measured. In total, 12 measures were derived from this task: 2 metrics (accuracy and RT) \* 3 stimulus types (words, objects, faces) \* 2 paradigms (delayed matching and surprise recognition). See [40] for a more detailed description.

### 2.3.2. Familiarity Decisions

Familiarity decision tasks were run for faces, objects, and words. For faces, participants were asked to decide whether a presented face was famous or not (80 trials in total). For objects, we used a 72-trial version of a well-studied object decision task [55], presenting line drawings of real objects and chimeric non-objects. Participants were asked to decide if the picture represents a real object or a non-object. For words, we used a lexical decision task with 60 trials, where participants were asked to decide whether the presented letter string represented a real word or a pseudoword. For all three familiarity decision tests, both accuracy and RTs for correctly categorized familiar items (famous faces, real objects, real words) were analyzed.

### 2.3.3. Naming of Familiar Items

Tests of picture naming (line drawings), face naming (famous faces) and word reading (regular words). For pictures and words, both accuracy and RTs for correctly named items are analyzed (a voice key was used for RT measurement). For famous faces, only accuracy is recorded as measuring RTs in face naming tasks is complicated by participants making other verbal responses than names (e.g., "it's that guy from the Parliament ... ").

### *2.4. Structural MRI: Lesions*

Structural brain imaging data were acquired from all subjects. Structural scans were acquired on two 3T Phillips Achieva scanners with 32-channel head-coils and a SENSE factor of 2.5 in London and Manchester. A high-resolution T1 weighted structural scan was acquired for spatial normalization, including 260 slices covering the whole brain with TR = 8.4 ms, TE = 3.9 ms, flip angle = 8 degrees, FOV = 240 × 191 mm2, resolution matrix = 256 × 206 and voxels size = 0.9 × 1.7 × 0.9 mm3. Automated outlines of the area a ffected by stroke were generated using Seghier et al.'s modified segmentation–normalization procedure [56]. Segmented images were smoothed with an 8mm full-width half maximum Gaussian kernel and submitted to the automated routines for lesion identification and definition modules using the default parameters. The automated method involves initial segmentation and normalizing into tissue classes of grey matter, white matter, cerebro-spinal fluid (CSF), and an extra tissue class for the presence of a lesion. After smoothing, voxels that emerge as outliers relative to the normal population are identified and the union of these outliers provides the "fuzzy lesion map", from which the lesion outline is derived. The generated images were used to create the lesion overlap maps.
