Incidental Detection of Hippocampal Sclerosis

The MR examination of medial temporal structures has recently been pushed to the limits of current technology, with some hope of uncovering the secrets of intractable epilepsy. The specific MR imaging features of one important source of seizures, mesial temporal sclerosis, have been well described. These include a small hippocampus with abnormal signal on T2-weighted scans, but many other findings in the temporal lobes and limbic system have been reported.

These careful investigations of brain anatomy have occurred without a critical examination of the hippocampus of nonepileptics to determine if these findings might be encountered as incidental findings in healthy patients, or at least in patients without epilepsy. This is not a simple task because high-resolution imaging in the coronal plane is essential for this diagnosis. In this issue of the AJNR, Moore et al (page 1609) report on the temporal lobe findings in 207 cases without known epilepsy in an attempt to determine the positive predictive value of the MR findings of mesial sclerosis.

In this retrospective study of patients with hearing loss, the authors identified two patients out of the 207 with abnormal hippocampal formations. In both cases, further investigation of the clinical history uncovered a history of epilepsy. The other 205 patients had normal medial temporal structures; ie, they found no cases with hippocampal abnormalities without epilepsy. The authors conclude that the findings of mesial sclerosis are “uncommon and significant.”

Although their observations are impressive, one should be aware of two important caveats. The first concern is how representative is the study sample? There can be little argument that this finding is uncommon. Of the patients referred from our epilepsy service with a suspected epileptogenic focus, MR studies are abnormal in approximately 25% of cases. With this experience in patients at risk, there can be little doubt that the occurrence of hippocampal abnormalities will be considerably less among nonepileptics. With an uncommon imaging finding, an important question is the number and composition of the study cases needed in order to reach a valid conclusion. It is essential that the cases in the study group are representative of the patient population (1). Ideally, this control group should include cases with a different disease process in the same anatomic location as well as patients with the same disease (epilepsy) but of nontemporal lobe origin. The outcome might have been different if the authors had decided to study patients with previous head trauma, near drowning, or herpes encephalitis, because all of these groups have a much higher likelihood of having temporal lobe abnormalities.

The cases of near drowning or anoxia are particularly relevant because these patients may have abnormalities limited to the hippocampus, attributable to a phenomenon referred to as “selective vulnerability.” It has been long recognized that specific regions of the brain might be injured with even brief episodes of hypoxia. Although vascular causes have been considered, the characteristic injuries seen in certain circumstances could only be explained by some pathophysiologic process unique to those cells. Selective vulnerability can be evident in many regions of the brain but is most commonly seen in the CA 1 sector of the hippocampus and the Purkinje cells of the cerebellum. This phenomenon has been attributed to the local release of excitatory neurotransmitters, and one likely candidate is glutamate, with secondary influx of calcium into the postsynaptic cells and subsequent injury. Another disease besides epilepsy and hypoxia that may selectively involve the hippocampus is limbic encephalitis, a rare paraneoplastic disease that in many respects resembles herpes encephalitis histologically.

The second caveat concerns the degree of statistical precision. A useful rule of thumb for estimating the upper bound of the 95% confidence interval around the probability of a rare event after N negative observations is 3/N (2). Using the data from this study, assuming that all of the patients with normal MR studies did not have epilepsy, the upper bound on the false-positive rate is 1.5% or 3/205; the lower bound on the specificity is 98.5%. Although this value is high, it must be considered in the context of the prior probability of disease. Among patients with suspected lesional epilepsy in whom the probability of finding disease is at least 10%, this specificity of 98.5% would yield a positive predictive value of at least 96%. In the general population, however, where the prevalence of epilepsy is approximately 8/1000, the positive predictive value could be no more than 40%. In fact, even at a prior probability of .008, a positive predictive value of 89% would require a specificity of 99.9%. To achieve this lower-bound estimate of 99.9%, Moore et al would have needed to review scans of 3000 subjects without finding a single hippocampal abnormality that was not associated with epilepsy. Although this would be a substantial undertaking, it would be of considerable interest to include some subjects with a history of anoxia in any such study group.

The authors are to be commended for their use of an existing data set to address this problem—a sort of scientific recycling. Their study provides solid evidence that the findings of mesial temporal sclerosis are significant in the clinical setting of epilepsy. The sample size, however, is neither large nor diverse enough to predict the true significance of these findings in the general population.