Visual agnosia in the era of behavioral and neural investigations

Abstract

Single case studies have long been used to provide insights into the mechanisms underlying normal cognition, including in the domains of memory, language and visuoperceptual function, and standardized testing has been a steadfast companion in such investigations. Experimental approaches designed to address specific hypotheses have also been conducted and analytic methods have been developed for the comparison of single subject data to a control group. However, a seismic shift has occurred in the last decade or two in which neuroimaging, primarily magnetic resonance imaging, has been added to the experimental toolbox. The question addressed in this article is whether, with these newer methodologies offering novel and previously unattainable evidence, single case studies have become obsolete. Here, in a single patient with integrative visual agnosia, tested repeatedly over three decades, behavioral, neuroimaging and joint behavioral-neuroimaging studies are described and their yield evaluated. Behavioral investigations have served to characterize the perceptual deficit well, and structural and functional neuroimaging data have furthered our understanding of the distributed circuit engaged in object recognition. However, imaging studies executed in concert with a behavioral task have offered more direct causal evidence, providing a more complete understanding of brain-behavior correspondences that goes beyond the sum of the parts. The conclusion reached is that the contribution of causal evidence from single cases remains a powerful methodology in advancing our knowledge of the neural basis of cognition.

1. Introduction

Visual agnosia refers to a diverse, albeit rare, class of neuropsychological disorders in which, following brain damage, an individual with normal premorbid visual perception is impaired at deriving the meaning of some or all categories of visual stimuli. The term ‘agnosia’, from the Greek ‘without knowledge’, can also manifest in the auditory (Miceli and Caccia, 2023) and tactile (Reed et al., 1996) domains, although these deficits are even less common than visual agnosia. Within the visual domain, there are various subtypes of agnosia: individuals with ‘object agnosia’ fail to recognize common everyday objects such as a lamp or a salt shaker (Riddoch and Humphreys, 1987), others with ‘dyslexia’ (or alexia) fail to recognize common written words (Starrfelt and Woodhead, 2021) while those with ‘topographic agnosia’ fail to recognize landmarks and scenes (Kim et al., 2015). Last, those with ‘prosopagnosia’ fail to recognize faces even those of close family members (Avidan and Behrmann, 2021; Barton, Cherkasova, Press, Intriligator and O’Connor, 2004).

The focus of the current paper is on the first of these perceptual impairments, ‘visual object agnosia’ (‘object agnosia’ for short) which can occur in isolation without any other concomitant deficit (Rumiati et al., 1994), although most cases have an accompanying prosopagnosia (Aviezer et al., 2007; Karnath et al., 2009). Descriptions of visual agnosia date back centuries and detailed descriptions of the behavioral manifestations have withstood the test of time and have helped advance our theoretical understanding of the mechanisms that give rise to normal object recognition (Behrmann and Nishimura, 2010; Marotta and Behrmann, 2004). While many previous studies have adopted measures from standardized psychological assessments generally designed for other purposes [such as the Boston Naming Test (Goodglass et al., 1983)] or existing sets of line drawings (Snodgrass and Vanderwart, 1980), test batteries such as the Birmingham Object Recognition Battery (Riddoch and Humphreys, 1993) or the Visual Object and Space Perception battery (Warrington and James, 1991) have specifically aided the diagnosis and evaluation of object agnosia.

In addition, experimental studies have been customised to address explicit hypotheses regarding the mechanisms underlying agnosia and, by inference, normal object recognition. Last, with the more recent advent of neuroimaging, studies have succeeded in shedding further light on the locus and extent of the neural alterations in agnosia. Here, data from a single case of visual agnosia, SM, collected over three decades, are used to evaluate the advantages of joint psychophysical and neural approaches, along with relevant single-case statistical measures (Crawford and Garthwaite, 2002; Crawford et al., 2006).

1.1. Manifestation of visual agnosia and defining symptoms

Visual object agnosia has typically been described in adults with acquired damage to visual cortex following etiologies such as a stroke or tumor (Albert et al., 1979; Arguin et al., 1996) or carbon monoxide toxicity (Goodale et al., 1991; Sparr et al., 1991). There are, however, also individuals with visual agnosia who sustained a neurological insult early in life (developmental object agnosia (Funnell and Wilding, 2011; Gilaie-Dotan and Doron, 2017); and even individuals with visual agnosia, often lifelong, in the absence of an obvious neural concomitant (Bate and Tree, 2016). Deficits in object recognition are also present in individuals with progressive degeneration resulting from posterior cortical atrophy (Mizuno et al., 1996), in those with semantic dementia (Rogers et al., 2003; Tree and Playfoot, 2015) and in individuals with neurodevelopmental disorders such as autism (Scherf et al., 2008). Here, we focus specifically on a single individual who had normal premorbid object recognition and who, subsequently, became agnosic in early adulthood following a cerebral insult.

Central to the definition of visual agnosia is the failure to access the meaning of visual objects in spite of normal sensory vision, language, semantics and intelligence. The agnosia deficit is manifest regardless of whether the visual objects are presented as a black and white line drawing, as a 2D image or photograph or as a real object, albeit to a lesser extent (Jankowiak et al., 1992) for real objects, likely because of the presence of additional cues such as color (Barton, 2011; Mapelli and Behrmann, 1997) and real world size (Holler et al., 2019). Importantly, too, individuals with visual agnosia demonstrate normal recognition of objects through modalities other than vision (touch, audition, and verbal description of its function), further attesting to the fact that the impairment does not arise from a difficulty in retrieving names (anomia) or in accessing the necessary semantic or conceptual information.

1.2. The nomenclature

The classical literature distinguished between two forms of object agnosia, ‘apperceptive’ versus ‘associative’ agnosia (Lissauer, 1890; Shallice and Jackson, 1988). Whereas the former refers to a deficit in deriving a stable percept from visual input, the latter refers to the specific failure to access stored semantic information normally evoked by visual stimuli (De Renzi, Scotti and Spinnler, 1969). A behavioral distinction often made is that patients with apperceptive agnosia, because of the severity of perceptual disorder, are usually impaired in matching, copying and drawing (but see other ideas on agnosia subtypes (Davidoff and Warrington, 1993; De Renzi and Lucchelli, 1993; Efron, 1968; Warrington and Rudge, 1995),). In contrast, patients with associative agnosia can match stimuli, copy visually-presented images, and draw from memory even though they may fail to recognize the stimulus and may even fail to recognize their own drawings shown to them at a later date (Bauer, 2006).

Consistent with the general difficulty in deriving a stable percept, apperceptive agnosia is associated with bilateral, diffuse damage to early visual cortex following, for example, hypoxia, carbon monoxide or mercury poisoning (De Renzi, 2000; Landis et al., 1982; Sparr et al., 1991). Associative agnosia, on the other hand, is a consequence of damage to later, downstream regions of inferotemporal cortex, including the fusiform gyrus, temporal gyrus, or even anterior portions of the temporal lobe. A more specific hypothesis is that object agnosia results from damage that affects the human lateral occipital complex (LOC) (James et al., 2003), a crucial area engaged in object perception as revealed by function MRI (fMRI) (Malach et al., 1995) and micro-electrode arrays (Decramer et al., 2019). There continues to be debate on whether unilateral or bilateral damage is critical for the associative form of agnosia, with some reporting that a unilateral right (Barton, 2008) or left (De Renzi, 2000) hemisphere lesion suffices, and others arguing for the necessity of bilateral damage (Goodale et al., 1991; Karnath et al., 2009; McIntosh et al., 2004) (see section 4.1 for possible resolution of this debate).

1.3. Longstanding challenges to the study of visual agnosia

Over the last few decades, there has been increasing pressure to resolve two aspects of visual agnosia. The first challenge concerns the absence of a clear understanding of the deficits in visual computation in such patients. Chief among the pressing issues has been the increasing realization that the terms ‘apperceptive’ and ‘associative’ are too general and that object recognition entails a greater number of subprocesses than just two. Much credit for acknowledging the coarseness of the nomenclature goes to Riddoch and Humphreys thorough their investigations of patient HJA (G. W. Humphreys, 1999; G. W. Humphreys and Riddoch, 1987; Riddoch and Humphreys, 1987). At the age of 61, HJA exhibited a severe perceptual deficit following the post-operative obstruction of the posterior cerebral artery. This obstruction resulted in bilateral ischaemia of the occipital lobe extending to the anterior temporal cortex, and affecting the inferior temporal gyrus, lateral occipitotemporal gyrus and the lingual gyrus. Riddoch and Humphreys (1987) postulated that patient HJA had neither an apperceptive nor associative agnosia but, rather, had a selective impairment in the integration of local elements into holistic shapes. They coined the term ‘Integrative agnosia’ to denote the fact that HJA was able to perceive parts of the stimulus but failed to combine them into an integrated whole. A similar pattern was reported in another patient with progressive multifocal leukoencephalopathy whose object recognition errors reflected just parts of the stimulus, for example, stating that a pencil is “a long, extended object with something knobby (pointing to the eraser) at one end”, and who often asserted that there were multiple stimuli in the display that contained just a single item (Butter and Trobe, 1994). The new ‘integrative agnosia’ label led to more fine-grained studies assaying the ability of patients to configure parts of an image into a gestalt and exploring the extent to which such patients retained any sensitivity to holistic representations (Aviezer et al., 2007; Behrmann and Kimchi, 2003b; Giersch et al., 2000; Ptak et al., 2014).

The second challenge has to do with delineating, with greater precision, the necessary and sufficient neurological basis for the deficit. While structural imaging (from CT scans to MRI) provided details of which regions were damaged, the advent of functional magnetic resonance imaging (fMRI), easy access to EEG, adoption of electrocorticography and even intracranial EEG (iEEG) (Cao et al., 2025) have offered novel high-resolution approaches to recording neural activity in typical and atypical observers. It is probably safe to say, at this point, that while we have some understanding of the necessary regions involved in visual object recognition, we still do not have a complete understanding of the sufficiency of such regions.

In this paper, data from studies of detailed behavioral assays and/or fMRI are presented for a single integrative agnosic patient, SM. The thesis offered is that joint brain and behavioral investigations of a single case reveal a whole that is greater than the sum of each of them and that this more direct causal evidence is perhaps best obtained in single patients with circumscribed lesions.

2. Case history

SM, a right-handed male, sustained a traumatic brain injury in a motor vehicle accident in 1995 at the age of 18 years. CT scans obtained soon after the accident indicated a contusion in the anterior and posterior temporal cortex of the right hemisphere (RH) accompanied by shearing injury in the corpus callosum and in the basal ganglia of the left hemisphere (LH) (see Fig. 1). Over the years, only the lesion in the RH occipitotemporal cortex was apparent and this remained stable on multiple structural scans acquired since then and on the imaging data presented in section 3 below.

Fig. 1.

A. Transverse and coronal MRI images show the focal lesion in the RH lateral occipitotemporal region. B. Error responses on the Boston Naming Test (Goodglass et al., 1983). C. Target scene. D. SM’s copy of scene.

SM recovered well after rehabilitation, aside from persistent visual agnosia and prosopagnosia. On one instance of testing of his object recognition abilities (which have also remained stable over time), SM identified only 171 of 260 line drawings on the Boston Naming test (controls 251/270) and required an average of 2140 msec to respond (controls 884.7 msec) (Behrmann et al., 2005). SM’s errors revealed that he had some visual information about the stimulus and did not possess semantic information about the object. His errors appear to arise from the imprecision in the perception of the image (see Fig. 1B). He also called a ‘cat’ a ‘hamster’, a ‘mountain’ a ‘tent’ and a ‘monkey’ a ‘rat’, again suggesting that he is able to derive some but not all of the visual information from the presented image (Behrmann and Kimchi, 2003b). As shown in Fig. 1D, SM can also copy but in a very laborious and segmental fashion and then often fails to recognize his drawings at a later date.

S.M. has corrected visual acuity to 20/20 and his vision is unremarkable in all other respects. He performs within the normal range on tests of low-level visual processing (judging size, length, orientation of stimuli, color, and motion). He has normal spatial frequency thresholds (Behrmann and Kimchi, 2003b) and has normal accuracy on tests that require matching of objects from different viewpoints or along a foreshortened axis (Riddoch and Humphreys, 1993), although he is significantly slower than controls on the last test presumably because he is using a feature matching strategy (see also Fig. 3).

Fig. 3.

Arrangement of dots into columns (A) and rows (B) for grouping by proximity and by similarity. (C.) Navon compound displays in which the identity of the stimulus is consistent (e.g. ‘S’ made of small ‘s’s) or inconsistent at the global and local level (e.g. ‘S’ made of small ‘h’s). D. (top) Mean RT and SE of control data and (bottom) Mean RT and SE for SM.

In addition to the deficit in object recognition, SM has prosopagnosia (Behrmann et al., 2005; Gauthier et al., 1999a). He scored 36/54 on the Benton Facial Recognition Test; 41 or higher is considered the normal range for adults while scores below 37 indicate impairment. The deficit also extended to the recognition of facial expression: in a discrimination task using face images morphed along a continuum between pairs of different facial expressions (for example, happy and surprise or anger and fear), SM performed significantly more poorly than three individuals with congenital (developmental) prosopagnosia whose performance was indistinguishable from that of a large group of thirty controls (K. Humphreys et al., 2007). Intriguingly, in an effort to train SM’s face recognition ability using the Greeble training paradigm that, in typical observers, leads to increased FFA activation (Gauthier et al., 1999b), SM learned to recognize some Greebles although he did not attain normal levels of performance and required substantially longer than control participants (Behrmann et al., 2005). He also showed improvement in recognizing common objects but, surprisingly, his performance on face recognition, albeit poor initially, was even more impaired following Greeble training. At termination of the training, on fMRI, the voxels identified in the fusiform gyrus that were face-selective prior to training were no longer so and were, instead, more Greeble-selective. These results raise a cautionary note for training studies in which, with limited neural capacity, tuning of neurons to the statistics of the input may result in competition for what might be limited representational space.

3. Identifying the psychological mechanisms underlying SM’s agnosia

The primary hypothesis entertained to explain SM’s object recognition behavior was that he was impaired at deriving a holistic representation of the input (perhaps also the basis of his prosopagnosia). We had surmised that SM, like HJA, had difficulty integrating the parts of an image, and three detailed experimental studies tested this hypothesis.

3.1. Deriving a shape from local elements

In this first study, to assess whether SM was able to derive a shape from local elements in an image, we used a prime-target paradigm in which a prime of varying duration was displayed followed by a pair of test images for same/different judgment. As shown in Fig. 2, the prime (comprised of few or many elements), which did not require a response, preceded a target pair of stimuli. The target pair which required a same/different judgement consisted of hierarchical stimuli made of local squares or circles of the same (diamond) or different (square) global shape as the prime. The key focus is on the test pairs requiring a ‘same’ response. In the Same Configuration (SC) trials, the test figures shared the global configuration with the prime but differed in the elements with the prime (both global diamonds) and in the Same Element (SE) trials, the test figures shared the local but not global configuration with the prime. We only analysed SC and SE trials. Participants were not required to respond to the prime and made a two-button same/different judgement on the test pairs. The stimulus-onset asynchrony (SOA) between the prime and target pairs varied with times including 40ms, 90 ms, 190ms, 390ms and 690ms, allowing us to assess, at different timepoints, whether the representation was more consistent with the elements or the configuration.

Fig. 2.

The prime-target matching paradigm. (left) Examples of the few and many displays showing the few- and many-element primes and the ‘same’ and ‘different’ target pairs. The ‘same’ target pair trials either share the same configuration (SC) with the prime or share the same elements (SE) with the prime. (right) RTs for correct ‘same’ trials only for SC and SE trials over the course of different prime durations on the x-axis for (A) controls and (B) SM.

Evident on the right in Fig. 2 panel A, for the few element displays, the controls took longer to make judgments about the SC than the SE test pairs, indicating that the elements, and not the configuration, were more rapidly available to them across almost the entire time course. For the many element displays, they did, however, respond more quickly to the SC than SE trials at 40 and 90 ms indicating that they derived the configural unit rapidly and then, only at later time points, individuated the disparate elemental components. SM, in panel B, like the controls, favored the SE over SC trials in the few elements condition. In contrast with the controls, however, for the multielement trials, he did not show the early and fast grouping of the configuration (SC) and, only with extended prime duration (around 390 ms and longer), did he then show the advantage for the SC over SE trials. Of course, his overall RT was much slower than that of the matched controls, further reflecting his laborious perceptual processing of the visual inputs, and his RTs are slower overall for the multielement display than few elements display, again likely a result of his sequential processing of the components.

Interestingly, SM had not entirely lost his ability to group elements together and could use simple Gestalt principles to group elements into rows or columns by proximity and similarity (of luminance). When shown a matrix of black (and white) dots, and asked to indicate whether the dots formed a column or a row (see Fig. 3A and B), SM performed as well as controls (all above 90 % accuracy), reporting ‘row’ and ‘column’ when the organization favored rows and columns.

SM also showed some residual sensitivity to global configurations made of multiple elements (as seen above at 690 ms for many element displays), as demonstrated in a task using compound or hierarchical (Navon) display at the local and global level (see Fig. 3C). In this task, a single stimulus was presented and the identity of the global shape (‘S’ or ‘H’) was or was not consistent with the identity of the local elements (‘S’ or ‘H’). On different blocks of trials, participants were instructed to report just the global identity or just the local identity using two response keys, one for ‘H’ and one for ‘S’. As shown in Fig. 3D, although SM was able to identify both global and local ‘S’ and ‘H’ stimuli with accuracy of 98.9 %, his median RT was 656ms for global inconsistent trials (464 ms for controls), and he showed a global advantage of 7.56 % (calculated for global inconsistent minus global consistent/100) (controls 9.1 %) (Behrmann and Kimchi, 2003a, 2003b). The RT difference between SM and controls is indicative of his prolonged efforts to derive the gestalt, although, interestingly, the global advantage was not different from that of the controls. That he showed a global advantage was surprising given the integration deficit but, with enough time to assemble the global shape serially, he was able to do so, confirming the results of the multi-element prime-target matching trials shown in Fig. 2.

SM’s residual sensitivity to object structure was also confirmed in an imaging study in which he (and another agnosic patient) viewed possible and impossible Escher-type images. Although the possible and impossible images were only minimally different (a single feature is changed), differentiating possible from impossible is rather easy for normal observers. In the magnet, blocks of the possible and impossible objects were shown and the participants performed a n-back task. Critically, in SM, the fMRI response in ventral, but not dorsal cortex, exhibited less sensitivity to object structure compared with the controls. As evident, regions along dorsal cortex showed sensitivity to object structure in spite of the ventral lesion, even though, behaviorally, SM continued to demonstrate an advantage for geometrically coherent possible over impossible objects (Freud et al., 2017). Together, these findings attest to residual global shape sensitivity perhaps undergirded by dorsal, rather than his damaged ventral, cortex.

The presence of a global advantage was also shown by integrative agnosic, HJA (G. W. Humphreys and Riddoch, 2001, 2006), who, in general, correctly processed local features of objects (G. W. Humphreys et al., 1985) but struggled to derive a more global description of objects; for example, shown a picture of a pepperpot, he stated that it is “a stand containing three separate pans; the top has a design on its lid”. SM’s profile was also similar to that of patient SE, who, like HJA, had a bilateral ischemic stroke, and who failed to identify global letters but did, in fact, show normal interference from the global to local identity when responding to local letters (Aviezer et al., 2007). Together these findings suggest that global shape information is not absolutely eliminated in these agnosic patients; rather, some sensitivity to the holistic configuration is preserved, but it does not appear to emerge naturally from perception and a laborious assembly process is likely invoked.

Based on these findings, we argued that SM was significantly impaired in apprehending a complex stimulus as a unified shape. These results also highlighted the fact that not all forms of perceptual organization are equally taxing, and grouping by similarity or proximity is simpler than deriving the integrity of a shape; similar conclusions are reached in studies of other patients including HJA (Giersch et al., 2000), NM (Ricci et al., 1999) and FGP (Kartsounis and Warrington, 1991). These results are also compatible with results showing that this simpler Gestalt-like grouping relies on areas like V1 and V2 (Lamme and Roelfsema, 2000) but that more complex configuration processing relies on more anterior cortical regions, likely including LOC which is lesioned in SM, as discussed further in Section 4 below.

3.2. Learning complex multipart objects

Clearly, SM has a deficit in deriving a global whole from especially from multiple element displays. While he does have access to the features, it’s their assembly that is challenging. As a further assessment of the assembly process, in the next study, we used novel objects, whose parts and their spatial relations determined category membership (shared overall shape) (Behrmann and Williams, 2007). The stimuli, labelled Fribbles (see Fig. 4 left), consisted of 3D rendered objects from six different categories with unique labels. Exemplars of a category were formed from a category prototype consisting of a main body (which was shared across a pair of categories) in concert with the locations of three attachments of the parts to the body. Thus, within a category, the main body and spatial relations of appendages remained the same but between one and three of the parts themselves varied (see last row for four DUVA exemplars). Two of the many experiments are reported here.

Fig. 4.

A) Exemplars from each of six species of Fribbles with the associated category label below, and, in the bottom row, four different exemplars from the DUVA category are shown with changes in parts varying from 1 to 4. B) Accuracy for controls and SM for same decisions and different decisions of exemplars from a single species that differ as a function of the number of parts. C) Accuracy for controls and SM when making decisions of exemplars from two different categories that do or do not share the same main body.

In one experiment, SM and matched controls had to decide whether a sequentially presented pair of Fribble exemplars were identical or not. On trials on which there was a difference, exemplars from the same category varied from each other by changing between one to four parts. As shown in Fig. 4B, SM’s accuracy was significantly lower than that of the controls across the board including for identical exemplars (data from Table 1 of Behrmann and Williams, 1997). His accuracy increased, however, as the two exemplars became less similar i.e. when the sequential Fribbles had three of four different component parts although he still performed more poorly than the matched controls even in the four-part difference trials.

In a further experiment, which directly addressed the deficit in integrating parts to form a whole, participants again viewed two sequentially presented Fribble exemplars and now had to respond via key press if they were members of the same category or not. To do this accurately, the spatial location of the appendages needed to be integrated and a more abstract representation of the whole derived. No single part (not even the main body as it is shared across species) suffices for the decision. Fig. 4C (data from Table 4 of Behrmann and Williams, 1997) shows that SM performed as well as the controls when the main body of the sequential exemplars differed, indicating different categories (for example, a SOGI versus a FIPO). However, when the main body was the same, for example, one exemplar was from SOGI and the second from KEZA, both of which have a long cylindrical purple main body, SM’s accuracy was significantly lower than that of the controls. Although the response should be ‘different categories’, when the main body was shared across two different categories, SM appeared unable to encode the parts and their relationship and, placed undue value on the main body and defaulted to this to guide his (erroneous) decisions.

3.3. Integration of two parts

We have demonstrated that SM has difficulty combining elements into a whole perhaps because he is unable to code the spatial relations of the parts. In this last behavioral study, we simplified the input and evaluated whether SM could learn simple shapes whose parts or their spatial relations varied. Participants learned to classify four target objects, all comprised of two simple three-dimensional volumetric parts or shapes and to respond via four-way button press (Fig. 5A top). The target-to-key mapping was reordered for half the controls). Participants then viewed the learned targets presented either from the studied viewpoint or a different vantage point, and used the four keys to indicate whether the object was a target or the space bar to indicate a distractor. Distractors were of two kinds: the parts of the target objects differed but spatial relations among the two parts remained the same (parts-change, PC; Fig. 5B) or the parts of the object were the same but spatial relations differed (relation-change, RC; Fig. 5C). PC distractors always included one part from a different target object. Controls learned the target classification and stimulus-response mapping in 2 blocks of trials, as did SM but he was given an additional block to consolidate his learning.

Fig. 5.

(left) A. Four target objects learned from a specific viewpoint. B. Part-changed distractors which incorporate one part from a different target object. C. Relation-changed distractors where the parts of the original object are retained but the spatial relation between the parts is changed. All of these objects were shown from different vantage points too. (right) Data from the No-Icons and Icons blocks. The control data shown in the first dark bar are collected under conditions equivalent to SM and the open bar reflects the results of the controls in the limited duration and masked condition.

The test phase included two blocks of trials. In the first, the No-icons test, the test item was shown in the center of the screen for a limited duration and participants responded which target it was using the keys 1–4 for targets or the spacebar for any other response. In the second block, the Icons test, the test item again appeared in the center of the screen but, in addition to this, the array of four targets, as in Fig. 5A, were shown at the top of the screen. These trials were akin to a simultaneous match-to-sample perception task, and were done to alleviate any memory load for SM, which might have adversely impacted his performance in the No-icons trials. The control participants completed one further version of the task under brief exposure duration and masking. This was done to test the hypothesis that if SM’s deficit was simply that he needed more time to perform correctly, the controls’ behavior under masked and limited durations might be more similar to that of SM’s results under the standard testing conditions.

The dependent measure of greatest interest was the false identification rate, i.e., how often SM would accept a distractor (PC or RC) as a target, as this would indicate whether SM failed to represent the parts or the spatial relations. The false identification rates are shown in Fig. 5 (right) for the No-icons and Icons blocks. Control participants made few false identifications and did not make significantly more when the duration of stimulus was brief and a mask was used (open bars on plot).

For the No-icons block, SM misidentified 21 % of the PC distractors (roughly equal to the mean of the controls in the masked brief version) and 52 % of the RC distractors as targets, which was significantly greater than that of the controls under either testing condition. While SM did make errors when the parts themselves changed, he made roughly double the number of errors when the parts remained the same as the target and only the spatial relations changed. Especially intriguingly, for the Icons block, controls made few errors and not significantly more on the masked, brief duration version but SM, who misclassified only relatively few of the PC distractors (~10 %), misclassified all of the RC trials (Fig. 5 right). This is surprising in that the presence of the Icons helped the controls. Dramatically, SM appeared to be captured by the test item sharing the same parts as the targets albeit in different spatial relations. Using various experimental controls, we were able to rule out the possibility that RC trials were simply harder than PC trials and these also allowed us to reject the possibility that SM’s errors occurred because he had encoded only one part of the two-part objects (see Behrmann and Williams, 2007). Instead, we suggested that, in the Icons block, the presence of the stimuli on the screen simultaneous with the test shape induced the use of a segmental and serial part-based strategy (as was seen in his copy of the beach scene, Fig. 1C) in which he matched the elements of the test with the visible targets. This was especially clear for the RC over PC distractors; for PC, he could reasonably accurately identify that the parts of the test stimulus and Icons differed, but, for RC, when the parts were shared, he failed to appreciate the spatial relations, and defaulted to using just the elements.

4. Summary

Three experiments, all of which targeted SM’s hypothesized inability to integrate parts of a visual display into a whole, resulted in several key findings: (i) SM had difficulty deriving a global whole from 2D displays as well as 3D images; he was especially impaired when there were few rather than many elements, when objects shared a main body and when the parts attached to the main body were similar; (ii) SM misidentified objects with the same parts as being exemplars of the same target even when the spatial relations between the parts differed; (iii) SM was able to group simple elements, such as dots, in a display into rows or columns by proximity, and other gestalt heuristics such as luminance, and showed some residual sensitivity to global shape; and (iv) SM struggled to learn objects with part-whole relations but, with extra time and laborious part-based assembly, he could do so, although not as well as the controls.

The engagement of a serial part-based strategy (perhaps obligatorily evoked) is consistent with reports of other cases of integrative agnosia, most notably HJA. The defaulting to processing the local elements is also considered to be a property, perhaps even causal, of prosopagnosia – in such acses in matching upside-down or inverted faces (which permit more part-based processing) is more accurate than matching upright faces (which relies on holistic processing which is defective) (de Gelder et al., 1998; Farah et al., 1995), and, thus, it may not be surprising that most integrative agnosic individuals are prosopagnosic, as well.

5. The contribution of neuroimaging

As noted in the introduction, the received wisdom has been that apperceptive and associative agnosia are differentiated by more posterior versus more anterior occipitotemporal lesions, respectively. Because many of the apperceptive cases became agnosic as a result of anoxia or carbon monoxide poisoning and inhalational intoxication, the lesions are typically diffuse and damage is to both grey and white matter across a large swath of posterior cortical regions bilaterally. In such cases, pinning down the necessary neural correlate of agnosia has been challenging. For example, structural scans of patient DF, who became agnosic after an anoxic episode, has revealed lesions in the bilateral occipital poles, ventral lateral-occipital cortex and LOC, including areas V2 and extrastriate cortex albeit to a greater degree in the RH than LH, along with a left parietal lesion and extensive cortical atrophy (James et al., 2003; Milner et al., 1991). Even in case of lesions not affecting early visual areas, the extent of the injury can also be large. Patient JS suffered a stroke, and had a bilateral lesion encompassing the fusiform and lingual gyri and isthmus of posterior cingulate gyrus, with further extension in the RH to the parahippocampal gyrus and cuneus (Karnath et al., 2009). Similarly, HJA had a bilateral ischemic lesion with bilateral damage to the inferior temporal gyrus, lateral occipital gyrus, fusiform gyrus and lingual gyrus (Riddoch et al., 2003). Understanding which of the affected areas play a more pivotal in agnosia is important but cannot be adjudicated based on cases such as these.

5.1. Functional neuroanatomy of object agnosia

SM initially presented with a rather more widespread lesion (see Section 2) on the first MRI scan post injury. In subsequent scans, much of this had resolved and the more recent MRI scans showed a rather circumscribed unilateral right lesion with a volume of 990 mm³ (Talairach-coordinates: +44, 46, 2), which coincided with functionally-defined area LOC (Konen et al., 2011). At first glance, the presence of a unilateral lesion offered a counterexample to most (perhaps all) other cases of associative agnosia described above. However, as revealed in our functional imaging study, the reduction of activation in the region of the unilateral lesion was mirrored by a reduction in activation in the homotopic region of the structurally intact hemisphere, essentially resulting in bilateral neural dysfunction.

As a mean of characterizing his neural deficit in greater detail, in this study, SM and matched controls viewed simple black and white objects such as a heart or a fleur-de-lis (all shown in 2D and also in 3D). The task was irrelevant but ensured participants were attending: they were required to count the number of changes in the colour of the central fixation point over a block of trials. To analyse the lesion carefully and evaluate responsiveness of the entire area, a grid was superimposed on the functional scan over the RH lesion and neighboring regions (see Fig. 6A and see lesion depicted in black ink), and did the same over the homologous regions in the LH. In each cell of the grid, the response profile was quantified and then compared between SM and a group of matched controls. Separately, we also compared SM against a single representative control to do a single case against single case comparison (see Fig. 6B). Three analyses were conducted with increasing stringency: (i) the mean percent signal change was calculated in response to viewing objects versus viewing a blank image, (ii) the proportion of activated cells (defined as significant activation of at least 50 % of the cell’s voxels) in the grid in response to an intact versus scrambled images, and, (iii) the adaptation effect (also called ‘repetition suppression’), calculated by subtracting the BOLD response to blocks of 16 repeated trials from blocks of 16 non-repeated trials.

Fig. 6.

A. Rectangular grid overlaid on retinotopic regions in SM. The posterior part of the grid was centered on the lesion site in the RH, and its mirror-symmetric location in the LH. Retinotopic areas and LOC are color coded. The cells overlaying the lesion site are shown in black. B. Rectangular grid overlaid on retinotopic regions in control participant, C1. This participant is shown in Figures C–E as well. C. Visually responsive activations in the grid of the group, SM, and C1. SM evinced activation to a visual display (versus blank image) over the same extent as the controls but with reduced amplitude. D. Object-responsive activations. The extent of regions in the grid being activated by intact over scrambled objects was significantly reduced in SM than in the group or C1.E. Adaptation effects (different minus same 16 trials in a block) were significantly reduced in SM versus the group and versus C1. LH = left hemisphere; RH = right hemisphere. (Adapted from (Konen et al., 2011)).

As can be seen in Fig. 6C, when viewing an object versus a blank image, SM showed widespread signal throughout the grid much like in the control group and the individual control although SM’s signal was significantly reduced in amplitude. However, both the intact minus scrambled objects condition (Fig. 6D) and the adaptation condition (Fig. 6E) revealed substantial reduction of signal both in the site of the lesion and in the surrounding cells of the grid. Most surprising, however, was that the cells of the grid sampling the LH homologous region were also reduced, with the profile of the LH homotopic regions closely mirroring those of the RH across the three testing conditions. This decrement in functional response in the structurally preserved hemisphere is assumed to reflect ‘diaschisis’ (from the Greek ‘dia’ meaning ‘in half’ or ‘across’ and ‘schizien’ meaning ‘to split) or the remote effects of the lesion. Diaschisis is assumed to arise from long-distance functional abnormalities which occur via afferent terminals in nodes which share a circuit with the affected region (Salvalaggio et al., 2020; Siegel et al., 2022), and this diaschisis implicates both hemispheres as potential neural bases of SM’s impairment. Although we delineated the functional neuroanatomy of SM’s agnosia, we cannot know definitively whether the RH LOC and/or the LH LOC was causally responsible for the agnosia. Also, SM was performing an orthogonal task and not object recognition and, thus, we are left to infer the relationship between the brain and behavior for object recognition, rather than being able to map the behavioral impairment simultaneously.

6. Joint behavior and imaging

One of the biggest boons afforded to neuropsychology over the last few decades is the opportunity to collect behavioral data and functional MRI using the same paradigm either simultaneously or asynchronously. Constraining behavior and the neural response under the same conditions permits the ascription of the behavioral deficit to particular brain region/s more directly.

6.1. Visual cortex activation in integrative agnosia during object recognition

In a behavioral experiment and large-scale fMRI mapping of visual responses in SM and matched controls (Freud and Behrmann, 2020), participants viewed objects which were either intact or parametrically scrambled, formed by chopping up the display into finer segments (4, 16, 84 and 256 parts) which were then reorganized (see Fig. 7 left top). Although participants viewed these images in the magnet and monitored changes in the color of the fixation point, accuracy and RT of recognition were acquired while participants viewed the same stimuli and reported the name of the displayed object outside the magnet. We also acquired a second scan from SM using the identical procedure 27 months later (and also from two controls) to obtain a measure of long-term reliability of the fMRI profile. It was important to understand whether, in the absence of intact recognition, his ability was deteriorating over time, as was reported in the case of HJA (Riddoch et al., 1999).

Fig. 7.

(Top left) The behavioral manipulation with an example of an intact stimulus on right and increasingly scrambled versions to the left. (Bottom left) Mean accuracy of object recognition as a function of scrambling using data obtained outside the scanner. Recognition accuracy decreased as a function of scrambling in controls and in SM. SM performed poorly, relative to controls, across all levels of scrambling excluding the most scrambled version (floor effect). (Right) Degree of slope in BOLD response across the scrambling manipulation calculated in every visually-responsive voxel. A. controls B. SM (see lesion marked with arrows). The darker red legend indicates object sensitivity or a steep positive slope with increasingly greatest response with less scrambling and the darker blue legend reflects a steep negative slope with greatest response to the most scrambled stimuli. (Adapted from (Freud et al., 2017)).

As is clear from Fig. 7 (left bottom), SM’s accuracy in object recognition is significantly poorer than that of controls at all levels of scrambling except at the maximally scrambled S256 at which point the controls performed at floor. Even for intact objects (full), there was roughly a 30 % difference in accuracy between SM and the mean of the controls. Also notable is the large drop in SM’s performance between intact objects and slightly scrambled S4 displays compared to controls, reflecting the fragile nature of his object recognition skills.

Because there was no statistical difference in SM’s fMRI profile across the two scan sessions conducted two years apart, attesting to the longitudinal stability of the fMRI results, we combined them for the analysis. With the fMRI data, object sensitivity was calculated for every voxel that was visually responsive (defined on independent data) by deriving the slope of the beta weights as a function of level of image scrambling (from intact to 256 parts) in that voxel. A positive slope in the beta-weight as a function of scrambling level reflects the increasingly stronger response as objects are less scrambled (shown in red), whereas a negative slope reflects the greater the response as object images are more scrambled (most edges, highest spatial frequency shown in blue). The colors in Fig. 7 (right) indicate the slope of the beta weight per voxel. Compared with controls (panel A), SM (panel B) had a significantly lower slope in the RH ventral cortex. Consistent with the presence of ‘diaschisis’ and, as shown in Fig. 6 (Konen et al., 2011), he also showed a significantly lower slope than controls in the LH ventral cortex in the vicinity of LOC. But, here, the results extended beyond those reported by Konen et al. (2011): the analyses also revealed lower beta weight slopes in voxels in the posterior parietal lobule of both the RH and LH in SM compared to controls, another example of diaschisis and the remote effect of the ventral lesion.

Together, the slope of the behavioral data, which confirmed the ongoing presence of agnosia and the reduction in beta weight slope in both hemispheres in both ventral and dorsal cortex, confirmed the consistency in the brain-behavior correspondence. These results also implicated a larger-scale network for object recognition than revealed thus far, and one in which LOC may be considered to be one of the preeminent areas for object recognition. Because a relatively local structural insult resulted in widespread altered function in ventral but also in dorsal regions, identifying the necessary and sufficient cortical locus for object recognition is not obviously possible (and, instead, a broader circuit may be necessary). It is also notable that large swaths of cortex, both ventral and dorsal, are also implicated in object recognition in the controls. Uncovering the widespread functional damage makes SM’s neural profile more consistent with those of other integrative agnosic patients whose structural lesions are much more extensive (Aviezer et al., 2007; Karnath et al., 2009; Riddoch and Humphreys, 1987). These data reveal that SM’s agnosic likely arises from what is essentially a bilateral functional lesion, even if the damage is unilateral. In the Discussion section below, we discuss possible but complementary differences in the dorsal versus ventral cortex contributions to object recognition.

6.2. Replication and signal mapping of diaschisis effects

The presence of diaschisis in SM (in the two fMRI studies above) offered us the further opportunity to uncover the neurophysiological properties of this remote loss of function, namely its temporal and spectral manifestation. To do this, we replicated the imaging study above but collected EEG data from SM (Fig. 8, red symbols) and matched controls (blue symbols) with the exact same scrambling paradigm (Chamanzar et al., 2025). EEG was recorded using 128 electrodes with a 10–5 standard high-density electrode montage and the ground truth of lesion was provided by the structural MRI study from Konen et al. (2011). The spectral decomposition of the EEG data into Delta, Theta, Alpha, Beta and Gamma bands was compared between SM and matched controls in three different regions of interest (ROI), as depicted in Fig. 8. In each ROI, the EEG signal was measured across the levels of scrambling and the slope of the signal derived for each frequency band (as was done for fMRI study above on a voxel-by-voxel basis).

Fig. 8.

Results of spectral decomposition of EEG signals for SM (blue) and controls red) in three regions of interest (ROI): dorsal cortex in pink, ventral cortex in green and SM’s lesion locus in black and its homologue in blue. In each ROI, the average shape sensitivity values was mapped for five frequency bands of Delta ([1, 4] Hz), Theta ([4, 8] Hz), Alpha ([8, 12] Hz), Beta ([12, 30] Hz), and Gamma ([30, 50] Hz). A significant decrement in shape sensitivity in SM, compared with controls, is observed across all frequency bands in the lesion ROI (all bands and p < .066 for Delta) and in the LH homologous region (but only in Theta band, p < .019) but also in dorsal and ventral regions (except for Delta and Theta bands in the RH dorsal, and Delta and Gamma bands in the left ventral cortex). (Adapted from (Chamanzar et al., 2025)).

The first ROI sampled was the locus of the lesion as determined on the structural scan from Konen et al. (2011). The data from the RH lesion (black rectangle) and from the homologous region in the LH (blue rectangle) are shown in the bottom row. The blue dots indicate the slope of the signal as a function of scrambling level for SM and the red dots are the slopes from each of the controls. As evident, there was a significant reduction in the slope of SM’s EEG object sensitivity in the lesion site in the Theta, Alpha, Beta and Gamma frequency bands (see asterisks denoting significant differences). There was also a reduction in slope in the homologous ROI of the intact LH in the Theta band. Intriguingly, during object recognition, Theta band is generally more restricted to and elevated in the inferior temporal cortex although during working memory tasks, the amplitude of Theta activity is lower but more widespread (Snipes et al., 2022).

We then examined the slope of the signal in ventral cortex (see Fig. 8 middle row). A reduction in slope was evident in Theta, Alpha, Beta and Gamma in the RH and in Theta and Beta in the LH. The last ROI investigated was dorsal cortex (see Fig. 8 top row). Last, consistent with the fMRI findings, the slope across scrambling levels of SM’s EEG signal was also reduced in the dorsal cortex of the RH in the same four frequency bands as in ventral cortex. Moreover, in the LH dorsal cortex, there was also a reduction in EEG signal in Theta bands, as was also so in the LH ventral signal. This frequency-dependent within and transcallosal diaschisis illustrates the ripple effect of the lesioned cortex on the function of disparate regions, thereby implicating a distributed bilateral object network both in ventral and in dorsal cortex.

Summary:

The advantage accrued over and above structural or functional neuroimaging alone by coupled imaging-behavior paradigms is that a more direct (and perhaps causal) link to the performance deficit can be uncovered. At the very least, the decrement in behavior for specific conditions can be matched by a decrement in neural activation: in both a fMRI and an EEG study, which used the same behavioral paradigm to tag object sensitivity, SM’s atypical neural profile corresponded closely to the drop in behavioral accuracy over the levels of scrambling of the object. Moreover, in both studies, the RH lesion had remote consequences, with homotopic ventral regions in the intact LH also evincing the decrement in neural signal associated with the corresponding behavioral decrement. Last, bilateral dorsal areas also showed reduction in object sensitivity. The derivation of the slope across levels of scrambling served as a useful signature to link together the behavioral and neural changes in SM in comparison with the control observers.

7. Discussion

The goal of this paper is to examine the nature and extent to which approaches such as detailed behavioral assessment and functional MRI, individually and jointly, have contributed to our understanding of visual object agnosia and of mechanisms underlying object recognition, more generally. The evidence presented is drawn from a single case, SM, who had premorbidly normal visual perception but became impaired following damage to the ventral visual cortex. SM’s perceptual deficit is well captured by the term ‘integrative agnosia’ as he fails to configure elements of a viewed stimulus into a holistic shape or configuration. In the course of studying SM over roughly three decades, we have obtained detailed but largely independent behavioral and neuroimaging data, and the insights from these two modalities have shed further light on the psychological and neural mechanisms of agnosia. Here, we argue, however, that examining SM’s behavioral profile under the exact same conditions collected during imaging informs our understanding of agnosia more than is possible by each approach alone. This coupled evidence permits us to identity more precisely the brain-behavior correspondences in object recognition in health and in disease.

SM’s lesion is circumscribed and falls within the RH lateral occipital complex (LOC, at the boundary of LO and pFs) (see Fig. 1), the region that shows the strongest activation in response to intact over scrambled objects in typical observers (Grill-Spector et al., 1999). This same region has also been damaged in other integrative agnosic patients, including AL (Ptak and Lazeyras, 2019; Ptak et al., 2014), HJA (Riddoch and Humphreys, 1987) and SE (Aviezer et al., 2007), and also implicated in apperceptive agnosic patient DF (James et al., 2003). We first reviewed evidence from behavioral investigations and from neuroimaging studies conducted independently. A first notable observation is that, over the three decades of testing, both the behavioral impairment and neural atypicality (beyond the acute period of the traumatic injury) have remained stable on anatomical scans and on functional activation measures (Freud and Behrmann, 2020). This is rather different from the well-known patient, HJA, whose long-term knowledge of the visual properties of objects deteriorated over time (Riddoch et al., 1999). There are several obvious differences between SM and HJA, including the fact that SM was decades younger than HJA at the onset of the brain damage and that his lesion was much smaller than HJA’s ischemic lesion. These, as well as other factors, may explain the differences in the preservation versus attrition of object representations. Suffice it to say that long-term visual representations acquired prior to SM’s injury appear to be preserved and he relies on these internal models in a top-down fashion while trying to make sense of or ‘problem solve’, as he says, his visual environment.

7.1. What we have learned from independent behavioral studies

Across three selected behavioral experiments, we characterized SM’s difficulty grouping elements into a configuration rapidly and efficiently. While he appeared unable to group sparse elements of a display easily, with enough time he was able to derive the global shape of objects (see Fig. 2). This was accomplished largely through a slow, sequential process, and he required prime durations at last ten times longer than those of matched controls to do so (Behrmann and Kimchi, 2003a, 2003b). In studies using 3D-rendered colored Fribbles drawn from six visually distinct categories, SM failed to appreciate the spatial relationships between the parts and the main body, leading him to err in deciding whether two Fribbles were the same or not (see Fig. 4). Also, SM appeared to default to a decision in which Fribbles that shared a main body was sufficient for determining whether two Fribbles belonged to the same category or not (this is erroneous as two different categories shared the same main body) (Behrmann and Williams, 2007). This failure to apprehend the spatial relations was replicated perhaps most clearly when simpler, two-part volumetric target shapes were used (see Fig. 5). Even though SM learned to distinguish four target objects, made of just two parts each, initially, he made 100 % false alarms, incorrectly classifying items whose two parts but not spatial relations matched one of the four learned targets. This was especially evident when the to-be-matched display appeared simultaneously with the presentation of the four learned targets on the screen (the Icon condition). Especially notable in all studies was the slow and laborious processing of the visual input with SM seemingly constructing an integrated visual representation part-by-part or even feature-by-feature (as also reported by Aviezer et al., 2007; Behrmann et al., 2006).

SM did, however, retain the ability to group displays perceptually based on similarity or proximity of the elements, a less taxing task than deriving the integrity of a shape (see Fig. 3). This retained ability to organize elements based on simpler Gestalt-type heuristics was also reported for patients HJA (Giersch et al., 2000), FGP (Kartsounis and Warrington, 1991) and AL (Ricci et al., 1999). Preservation of these simpler forms of perceptual organization is also consistent with SM’s more anterior lesion site: simpler grouping relies on areas like V1 and V2 (Lamme and Roelfsema, 2000) which are preserved in these patients but more complex configuration processing relies on more anterior ventral regions, which are damaged. SM, like some of the other patients too, also retained some semblance of sensitivity to object structure, for example, showing an advantage for the global identity of Navon hierarchical stimuli and also differentiating between possible and impossible objects (which differed on a single feature that altered the 3D geometry). With enough time, he succeeded in assembling a global shape albeit with great difficulty.

7.2. What we have learned from independent neuroimaging studies

Neuroimaging of patients with visual agnosia has, to date, largely consisted of describing the results of a structural MRI scan to localize the site of the lesion, usually collected as part of the clinical evaluation. For the most part, the lesions of integrative agnosic patients are relatively widespread, implicating regions on the lateral and inferior surfaces of the temporal lobe, but often extending anteriorly as well. Given the extent of the lesion, it has been hard to pinpoint historically a specific subregion that necessarily and sufficiently is causal in integrative agnosia, although a LOC lesion in either hemisphere [SM in the RH and JS (Ptak et al., 2014) in the LH] is sure to be a key, but perhaps not the only, component.

Functional MRI offered new opportunities to track the neural basis of the perceptual impairment. In one study, we measured the BOLD activation while SM viewed 2D and 3D images of objects projected into the magnet (see Fig. 6). We analysed the data under three different conditions, incrementing in complexity (Konen et al., 2011), including measuring activation in response t visual objects in contrast with fixation, contrasting responses to objects compared with scrambled images and quantifying adaptation using a repeat/nonrepeat design to measure suppression of BOLD to repeated trials. Using a grid superimposed on the lesion and its homotopic region in the LH enabled us to document visual activation to objects versus fixation in the penumbra of the RH lesion. In the other two experimental conditions, activation was limited in magnitude and extent compared to that of the controls. Most surprising, this latter aberrant profile was observed in the homotopic regions in the intact LH, reflecting the remote consequence of cortical damage.

The same finding of signal reduction in the homotopic area of the preserved hemisphere was reported for patient AL, who had integrative agnosia after a LH LOC lesion (Ptak and Lazeyras, 2019). Also, in AL, the functional connectivity (correlation of the time series of the BOLD profile) was reduced between the damaged LH and the intact RH medial/lateral occipital cortex confirming that focal damage to LOC can have widespread ramifications in bilateral occipito-temporal cortex. The findings of diaschisis and the mimicking of a bilateral lesion suggest that bilateral temporal representations may be necessary for shape perception. It is the case, however, that unilateral engagement is inadequate; in individuals with just a single hemisphere (after childhood hemispherectomy for drug-resistant epilepsy), category-selective activation and the adaptation profile in the preserved hemisphere was equivalent to that of controls, but accuracy of recognition measured was statistically inferior to that of controls. Together, the findings led to the conclusion that two hemispheres are not only better than one but are, in fact, necessary for normal performance (Robert et al., 2024).

7.3. What we have learned from combined behavioral and neuroimaging studies

While much has been gleaned from each of the separate behavioral and neuroimaging investigations, two key findings emerged when the same object recognition task was completed behaviorally and in the magnet. The first result is that the diaschisis was confirmed: the reduction in activation around the lesion site was mirrored qualitatively and quantitatively in the homologous ventral cortical region, relative to controls. This cross-hemispheric diaschisis was also evident in a task in which levels of levels of object scrambling was parametrically manipulated in fMRI (see Fig. 7) (Freud and Behrmann, 2020; Freud et al., 2017) and in EEG (see Fig. 8) (Chamanzar et al., 2025). The same decrement in neural activation and accuracy in the behavioral profile, across the levels of scrambling of the display objects, was uncovered both in the RH lesion site and in the LH homotopic region.

Over and above this interhemispheric diaschisis, however, was the concomitant drop in object sensitivity across levels of scrambling bilaterally in dorsal cortex in both the fMRI and EEG portion (and also see (Freud et al., 2017a) for further demonstration of dorsal and ventral cortex limitation in SM). The parallel behavioral and neural signatures have brought into registration the behavior-brain correspondence, and have provided converging evidence of a neural circuit with multiple nodes, at least in dorsal and ventral cortex in both hemispheres. Damage to one node affects signal transmission to other nodes in the network. Thus, while the LOC lesion may be pivotal, it is clear that multiple regions are activated in response to viewing objects and further parcellation of the various functional roles in this circuit is needed.

One specific future direction that pertains to SM, however, involves the possibility of changes in structural connectivity as a result of the lesion. To be specific, one perhaps puzzling outcome of the studies with SM is that the lesion is relatively limited, (covering only a small portion of LOC right at the boundary of LO and pFs), yet, the consequences implicated a broad set of object-selective regions. A possible explanation is that the lesion at the boundary of LO and pFs interrupted the propagation of signals to other regions. If so, it might not the lesion per se that is at fault but rather, the obstruction of signal propagation through this region serves as the basis for the diaschisis and remote effects. Examining both structural and functional connectivity seem like very obvious and, hopefully, telling next steps.

7.4. Dorsal cortex and shape representations

The activation of dorsal cortex during object recognition in typical observers and its reduction of object sensitivity in SM uncovered in both the fMRI and EEG studies is surprising (although less so as more studies report similar results, see below). One possible explanation is that the observed dorsal activation is simply the cascade of ventral signals to dorsal regions via the substantial white matter connectivity between them (Baizer et al., 1991; Yeatman, 2024; Yeatman et al., 2014). This is likely not the sole explanation and we demonstrated that there may, in fact, be independent object representations computed dorsally. The sensitivity to object structure (possible versus impossible) in the dorsal cortex of patients with ventral lesions, as measured by fMRI, fell within the range of the controls’ sensitivity. Moreover, this dorsal activation was still present even when any residual ventral cortex sensitivity in the patients, including in SM, served as a covariate (Freud et al., 2017b). Of relevance here is that, in addition to SM, many other integrative agnosic patients with very broad ventral lesions also exhibit ongoing sensitivity to global shape. This is true of HJA (G. W. Humphreys et al., 1985), SE (Aviezer et al., 2007) in his performance on global-local tasks, and AL who retained intact global and local processing of 2D and 3D object structure (Ptak et al., 2014). Even an apperceptive agnosic patient with a large ventral deficit retained some sensitivity to the integrity of object structure (Holler et al., 2019).

But if dorsal cortex contributes to shape perception, then patients with parietal lesions should be agnosic. While this is clearly not so, some studies have reported causal effects of parietal damage on perception. For example, in non-human primates, reversible inactivation of posterior intraparietal area resulted in reduced activation of ventral cortex and, specifically, a perceptual deficit in depth-structure categorization (Van Dromme, Premereur, Verhoef, Vanduffel and Janssen, 2016). Also, single-cell recording of the very same area revealed selectivity for the 3D orientation of planar surfaces and for the structure of depth derived from disparity (Alizadeh et al., 2017). In related fashion, one proposal has argued in favor of a dorsal, global shape representation that emerges from a description of the spatial arrangement of a basis set of local features which are ventrally mediated (Ayzenberg and Behrmann, 2022). Relatedly, multivariate decoding of high-density EEG acquired while participants viewed objects revealed that dorsal cortex has temporal precedence and predicts the response of ventral cortex rather than the other way around (Ayzenberg et al., 2023).

The more rapid signal propagation is consistent with the faster signaling in the magnocellular pathway through parietal cortex (Bar et al., 2006; Merigan and Maunsell, 1993) and, potentially, the exploitation of rapidly propagated low spatial frequency signals (G. W. Humphreys et al., 1992). Many other theoretical accounts of dorsal cortex function have also been offered including the possibility that the geometric and spatial attributes represented by dorsal cortex inform the perception of potential actions on objects and, as a by product, are available for perception (Freud et al., 2020). Together, these proposals implicate both dorsal and ventral cortex, potentially with complementary representations of more global versus more local feature computations. Future studies will undoubtedly assist in uncovering and formulating more precisely the individual contribution of dorsal cortex to object recognition.

7.5. The special sauce of single case studies

The key argument offered in the current paper is that single case studies are not obsolete and that, notwithstanding the arrival of powerful methodologies that permit access to human brain data in unprecedented ways, single case studies, especially with coupled behavioral and imaging data, contribute substantially to our understanding of the mechanisms mediating complex cognition. The ability to conduct in-depth studies is a further major advantage of studying a single case. Single-case designs allow multiple, fine-grained behavioral assays that probe the deficit in depth and, in this case, do so over longitudinal time and provide unique insights into underlying mechanisms, especially when behavior and imaging investigations are tightly coupled. One possible future direction to increase the contribution from studies of patients can be borrowed from structural imaging of patients.

This is not to claim that large scale studies are not informative. Neuropsychological studies with multiple participants either in case series format (Behrmann and Plaut, 2014) or en masse as a group (Rice et al., 2021) have substantially furthered our understanding of behavior-brain relationships. It is worth noting, however, that such studies, with some exceptions, typically rely on a small number of behavioral measures. Of course, adding further comprehensive behavioral batteries and targeted experiments would strengthen their explanatory power.

Imaging studies of typical observers have also contributed importantly to our understanding of visual behavior (Grill-Spector et al., 1998; Weiner et al., 2018; Yan et al., 2024), and the current access to large datasets from typical individuals such as the Human Connectome Project (Glasser et al., 2016), continues to open new vistas for research. Although many have argued that fMRI only provides correlational findings and is unable to address causal relationships, others have suggested that imaging studies can provide causal information about the relationship between behavior and brain events, but they still cannot offer causal certainty (Weber and Thompson-Schill, 2010). Either way, the impact of an absent region of cortex on subsequent function and behavior is indisputably causal.

Very few group studies have a functional imaging component; it is challenging (although not impossible) to conduct fMRI with a large-ish sample of patients but many neuropsychological disorders are simply too rare to be aggregated into a group. Obviously, scanning a large-ish group of patients, where possible, would help address generalizability of the results from single cases. Many studies have calculated the frequency count of overlapping lesions per area in patients who share a deficit of interest. These types of functional prevalence maps help triangulate on the location or perhaps even the network associated with the affected behavior (for example (Geva et al., 2021; Lorca-Puls et al., 2021), and can be used to make predictions about prognosis (Hope et al., 2013). Duplicating this approach in fMRI might also bridge brain and behavior by examining patients with a similar deficit even if they have non-overlapping lesions or different etiologies. Drawing up a map of reduction of activation in varying regions of interest or, better yet, in voxels and in whole brain measurement, may nudge us further in understanding the contribution of an area or many areas to the observed deficit.

8. Conclusion

Neuropsychological case studies have played a storied role in the history of cognitive neuroscience. Their continued contribution might appear to be on the wane given the introduction of new methodologies that offer extraordinary insights into the neural correlates of behavior in typical individuals. But this has not happened and, if anything, the behavioral investigations of patients in parallel with imaging studies have offered new opportunities to learn about causal mechanisms underlying behaviors such as perception, memory and language.

Here, in a single patient, SM, studied over more than three decades, the contribution of hypothesis-driven behavioral experiments and of detailed structural and functional imaging have helped advance our understanding of object recognition in health and its breakdown in an individual with visual object agnosia. Over and above the independent study of behavior and the neural profile, combining these methods with matched testing conditions and powerful analytic approaches has opened a new vista on object agnosia. Specifically, the apparent engagement of a distributed network of regions bilaterally reveals a neural circuit of ventral, but also of dorsal, cortex activated during the process of object recognition. The open questions now include the relative contribution and necessity and sufficiency of each region in this circuit. And while we have some glimmers into the temporal dynamics of these computations, much remains to be revealed. Rigorous case studies of individuals like SM will continue to be useful for elucidating the neural basis of behavior and for offering unique, causal evidence in the service of furthering our knowledge of the neural underpinning of complex human behavior.

Data availability

No data was used for the research described in the article.

Data and code availability

Data from behavioral studies are available by request from author.

For Fig. 7: Materials and data for the study are available at https://figshare.com/collections/The_large-scale_organization_of_shape_processing_in_the_ventral_and_dorsal_pathways/3889873.

For Fig. 8 EEG: he anonymized raw EEG dataset (for both controls and SM) and MRI scans (only for SM) used in this research can be shared upon request.