This investigation was conducted to test the proposition that generation of diagnostic hypotheses evoked by the immediate visual recognition of clinical signs engages neural systems that are recruited when naming objects in everyday life. The results support this hypothesis by showing very significant and similar activations in a circumscribed set of distributed cortical regions when naming radiological lesions and animals in the same context. However, diagnosing lesions was cognitively more demanding and associated with more activation in higher order cortical areas.
Higher mean RTs, error, and hesitation rates suggest that, on average, diagnosing lesions was more difficult than naming animals in our experimental setting. This could be related to the visual characteristics of lesions and/or the fact that low-frequency (in everyday language) lesion names are more difficult to recall, compared to high-frequency animal and letter names 
. Lexical semantic associations, i.e. being aware of words/concepts other than that vocalized, were more frequent while diagnosing lesions (), as compared to the other two categories, also indicating greater cognitive demand related to the selection of appropriate names in this particular task.
A relevant aspect of our results was that generation of diagnostic hypotheses can be very fast; the mean RT to diagnose lesions was 1.33 seconds. It is possible that the training before the fMRI experiment contributed to this performance. But very rapid identification of lesions (<1 second) has already been reported in radiology studies 
An important finding was that radiologists were able to cogitate differential diagnoses during the 3.5 seconds of a trial. A similar awareness of alternative names of other animals was reported while naming animals. In a few cases even letters evoked words to subjects (). Participants were not instructed to make differential diagnoses or to think about alternative names for animals during the task. These results are compatible with a fast and automatic semantic association process, in which the recall of a diagnosis or a name of an animal occurs with the concomitant activation of semantically related concepts 
. Clearly, a formal and definitive diagnosis can not be made in seconds but its core cognitive process, the generation of diagnostic hypotheses which were the names of lesions in our study, is crucial for a final correct diagnosis 
Diagnosing lesions in radiological images can be conceptualized as a process that is similar to localizing and naming objects in a scene 
. There are several fMRI studies of location and recognition 
or naming objects 
but we are unaware of studies in which all of these tasks are combined. We found very significant brain activations associated with the localization, recognition and naming of targets in radiological images, be they lesions, animals, or letters (). Activations were greater in the left inferior frontal sulcus and posterior cingulate cortex for naming lesions than animals and also greater for naming animals than letters ( and ). When lexical semantic associations, more frequent for lesions (), were taken into account in an analysis of covariance the difference of activations between those regions was no longer significant (at a corrected level). All the regions activated in the present study have also been reported in object naming in other studies 
, therefore there was no indication that the participating radiogists were naming animals differently from laypeople.
In agreement with our findings, activation in the left inferior frontal sulcus was reported in two fMRI studies of semantic verbal fluency - generating and vocalizing associated words in response to a word - in which there was greater lexical semantic demand in contrast to the comparison task, reading aloud 
. This is consistent with increased cognitive control when it is necessary to make a choice between synonymous or competing concepts, e.g. synonym words or differential diagnoses in our study, respectively 
There were regions more activated by naming animals relative to diagnosing lesions. These regions are usually associated with visual processing and recognition of stimuli; namely, posterior fusiform cortex and posterolateral occipital cortex ( and ) 
. We believe that differences in the visual characteristics of the stimulus categories we used are responsible for the observed differences in activations in these cortical areas.
The cognitive mechanisms underlying medical diagnosis have been studied with different conceptual strategies 
. One important approach relevant to the present study considers it a classification process similar to categorization in everyday life and several authors investigated diagnostic processes with different categorization models 
. These studies explored several aspects of the cognitive psychology of diagnostic reasoning but did not include a comparison of medical diagnosis tasks with, e.g., categorization of objects.
Interestingly, from a historical point of view, similarities between the classification of diseases and living creatures have been suggested in the past. In the XVIIth
century Thomas Sydenham in his influential definition of diseases proposed to conceptualize them as specific (ontological) entities similar to plant species 
. Influenced by Sydenham ideas, Boissier des Sauvages created a classification of diseases, nosology, based on taxonomic principles used in botany and his approach was followed by other phyisicians in the XVIIIth
century, including Carl Linnaeus 
Instead of using categorization, a concept with different meanings and diverse and conflicting models 
, we chose naming as a conceptually more prudent and descriptive approach. Picture naming has been a model extensively used in cognitive psychology 
, and functional neuroimaging 
to investigate how objects are recognized and named.
Some cognitive processes underlying medical skills have been studied with fMRI: one study investigated the neural substrate of visuo-spatial skills in surgery residents 
and the other compared brain activations in radiologists versus lay participants while viewing X-ray images 
. However, the neural basis of the medical diagnostic processes per se has not been investigated before.
Our experimental design was not planned to assess the generation of differential diagnostic hypotheses. Taking into account the limitations of the cued retrospective recall employed in the study, the conclusions resulting from the lexical semantic associations data need to be replicated using other methodological approaches.
In contrast to naturalistic and observational studies, planned experiments are by definition artificial and reductionistic due to the need to limit and control independent variables. Radiologists in their usual practice do not usually vocalize their diagnostic hypotheses as they come to their awareness. Conversely, we do not habitually vocalize the names of objects as we recognize them in our everyday life.
Radiologists customarily verbalize their diagnosis using more than one word, e.g. pleural effusion in right hemithorax, in contrast to one-word names for animals. This difference in length of responses could introduce an important confound variable 
. To circumvent it, we trained the participants to preferentially use one-word names to diagnose lesions (see Methods and Appendix S1
for details). Probably as a consequence of this experimental stratagem the complete names of the lesions were reported to came to the awareness of participants and were considered competing lexical-semantic associations (see Results
Under the blanket rubric ‘medical diagnosis’ there are different cognitive tasks and processes. Considering the case in point of radiological diagnosis: The immediate recognition and diagnosis of an obvious lesion probably recruits different neurocognitive processes as compared to the diagnosis of a subtle and ambiguous alteration with complex differential diagnoses requiring a detailed examination of the radiological image. To create our experimental design we had to limit the scope of the investigation and the conclusions of our study are restricted to the diagnosis of lesions that are immediately identified and diagnosed. It will be important to replicate these results with other approaches, e.g. electrophysiological methods such as electroencephalography or magnetoencephalography. The conceptual hypothesis also needs to be tested in other medical specialties in which diagnosis is strongly based on visual clinical data, e.g. dermatology.
This study is an attempt to investigate the brain mechanisms subserving medical diagnosis. We have demonstrated that differential diagnoses can be automatically elicited in a time frame of seconds in response to clinical signs. Our results support the hypothesis that a process similar to naming things in everyday life occurs when a physician promptly recognizes a characteristic and previously known lesion. In our experimental model, the diagnostic task was cognitively more taxing; more activation in higher order cortical areas was plausibly associated with demands related to the selection of appropriate names as compared to the control task.
The importance of non-analytical reasoning in medical diagnosis has been increasingly stressed 
. Our study is a contribution to the understanding of its mechanisms. There are recent reviews proposing the application of the knowledge acquired in neuroscience to improve medical education methods 
. An implication of our results is that information obtained from cognitive neuroscience studies on the recognition and naming of objects can be brought to bear on the improvement of diagnostic expertise in the visual domain. In addition, the conceptual hypothesis and the methodological approach described in the present investigation may open new ways to develop studies in medical diagnosis.