Endocast

The study of endocasts has its origins in the early 19th century. Georges Cuvier is credited with the earliest recognition and description of a natural endocast, when he described and figured the internal mold of the cranial cavity of an Eocene mammal from Montmartre, Paris, in 1804. In his Recherches sur les Ossemens fossiles (1822), Cuvier also noted a natural endocast visible in a fossil bird specimen (an 'Ornitholithe') from the same locality. These observations laid the conceptual groundwork for what would become paleoneurology.

In the late 19th century, Othniel Charles Marsh contributed significantly to the field by producing artificial endocasts of dinosaurs and fossil birds. His 1880 monograph on Hesperornis and Ichthyornis and his 1884 monograph on Dinocerata included some of the earliest physical endocasts of non-avian dinosaurs and large fossil mammals, respectively. Richard Owen (1871) produced the first figured endocast of an extinct bird, the giant moa Dinornis.

The field of paleoneurology was formally established by Ottilie ('Tilly') Edinger (1897–1967), a German-American paleontologist. For her doctoral dissertation (1921), Edinger studied the endocast of the Triassic marine reptile Nothosaurus, recognizing that the internal surfaces of fossil braincases preserve an imprint of the brain. In 1929, she published Die fossilen Gehirne ('Fossil Brains'), a comprehensive synthesis that outlined the methodology and central questions of paleoneurology. This work is considered the founding document of the discipline. Edinger demonstrated that endocasts, while not direct representations of brains, preserve sufficient information about brain shape and size to permit meaningful evolutionary comparisons across geological time. Her later career at Harvard's Museum of Comparative Zoology continued to shape the field through her 1975 annotated bibliography Paleoneurology 1804–1966.

Natural endocasts (Steinkerne) form when sediment fills the cranial cavity of a buried skull and subsequently lithifies. As the surrounding bone erodes or is removed, the stone cast of the internal cavity is exposed. These are relatively rare and depend on specific taphonomic conditions.

Artificial physical endocasts are created by pouring latex or other molding materials into the cranial cavity through the foramen magnum, extracting the mold, and using it to produce a plaster cast. This technique, described by Gratiolet (1858) and refined by later workers such as Radinsky (1968), was the standard approach through most of the 20th century.

Virtual (digital) endocasts are generated from CT, micro-CT (μCT), or synchrotron scan data. The endocranial space is digitally segmented from the serial sections using software such as Avizo, Amira, VGStudio MAX, 3D Slicer, or OsiriX. Virtual endocasts have become the dominant method since the late 1990s, offering non-destructive access to specimens, higher resolution, and the ability to partition the endocast into functional neuroanatomical regions. The first digital reconstruction of a non-avian theropod endocast was produced by Knoll (1997), and Brochu (2000) produced a digitally-rendered endocast for Tyrannosaurus rex.

The reliability of an endocast as a proxy for actual brain morphology varies significantly across taxa and across ontogeny. This relationship is often expressed as the brain-to-endocranial cavity (BEC) index. In highly encephalized groups—particularly adult mammals and birds—the brain nearly fills the endocranial cavity, and endocasts closely approximate brain volume and shape. Iwaniuk and Nelson (2002) demonstrated this relationship across 82 bird species. Watanabe et al. (2019) provided quantitative data showing that in the domestic chicken (Gallus gallus), the brain occupies approximately 60% of the endocranial space in neonates but over 90% in somatically mature individuals, with strong volumetric correlation (R² > 0.92). In the American alligator (Alligator mississippiensis), by contrast, the trend is reversed: the brain fills approximately 90% of the endocranial space in perinatal individuals but may occupy as little as 29–32% in large adults, due to indeterminate growth and negative allometry of the brain relative to body size.

Watanabe et al. (2019) further demonstrated that brain–endocast shape correspondence is not uniform across brain regions. The cerebrum and optic lobes show higher correspondence than the cerebellum and medulla, where the dorsal longitudinal venous sinus and other dural structures introduce greater deviation. This finding has direct implications for interpreting endocasts of extinct archosaurs, including non-avian dinosaurs.

Endocasts have been central to inferring the neuroanatomy and potential cognitive abilities of non-avian dinosaurs. Harry Jerison's Evolution of the Brain and Intelligence (1973) introduced the concept of the encephalization quotient (EQ)—the ratio of actual brain size to expected brain size for a given body mass—and applied it to dinosaur endocasts. Endocast studies have revealed that some theropod dinosaurs, particularly maniraptorans close to the origin of birds, had relatively large brains. Balanoff et al. (2013) used high-resolution CT to compare cranial volumes of extant birds, Archaeopteryx lithographica, and closely related non-avian coelurosaurs, showing that the enlarged, bird-like brain evolved before the origin of flight itself—representing an exaptation rather than a direct adaptation for aerial locomotion.

Among specific dinosaur taxa, endocasts have provided significant information. The endocast of Tyrannosaurus rex, first digitally rendered by Brochu (2000) and further studied by Witmer and Ridgely (2009) and Hurlburt et al. (2013), revealed large olfactory bulbs suggesting acute sense of smell, as well as a cerebrum of moderate relative size. The endocasts of troodontids such as Zanabazar junior and Saurornithoides mongoliensis are among the largest relative to body size in non-avian dinosaurs, approaching values seen in some modern birds.

In 2023, neuroscientist Suzana Herculano-Houzel published a study arguing that Tyrannosaurus rex may have possessed neuron counts comparable to those of modern baboons (over 3 billion neurons in the telencephalon), implying primate-level cognitive abilities. This claim was based on extrapolations from endocast volume and neuronal scaling rules derived from extant birds. However, Herculano-Houzel's study assumed that endocast volume equaled brain volume in T. rex, an assumption challenged by most paleoneurological evidence indicating that theropod brains did not fully fill the endocranial cavity. In 2024, a large international team including Hady George, Darren Naish, Cristián Gutierrez-Ibáñez, and others published a detailed rebuttal, arguing that the original neuron count estimates were based on flawed assumptions about brain–endocast ratios, inappropriate use of avian scaling laws for non-avian dinosaurs, and oversimplified cognitive inferences. The rebuttal concluded that T. rex cognition was likely 'reptile-like' rather than primate-like. This debate highlights the critical importance of accurately understanding brain–endocast relationships when making cognitive inferences from endocasts.

The endocast approach extends beyond the brain cavity. Digital endocasts of the inner ear (bony labyrinth), including the semicircular canals and cochlea, are widely used to infer vestibular sensitivity, locomotor agility, hearing capabilities, and habitual head orientation in extinct vertebrates. Endocasts of nasal cavities, neurovascular canals, and pneumatic sinuses also provide information about respiratory function, thermoregulation, and sensory systems. These applications have been particularly valuable in studies of pterosaurs (Witmer et al., 2003), early mammals (Rowe et al., 2011), and dinosaurs.

The construction of digital endocasts involves several key decisions that can affect outcomes. Scanner selection ranges from medical-grade CT (widely available but lower resolution) to micro-CT and synchrotron scanning (higher resolution, better suited for dense fossils). Segmentation—the process of digitally isolating the endocranial space—can be performed using threshold-based (more objective) or freehand (more subjective) methods, or a combination of both. Best practices, as outlined by Balanoff et al. (2016), recommend approximately 1,000 slices through the braincase and careful definition of boundaries at foramina and fenestrae.

Common artifacts include beam hardening (differential filtration of polychromatic X-rays causing uneven grayscale values), beam starvation (insufficient X-ray penetration of dense specimens), and ring artifacts (detector calibration errors). These can complicate segmentation, particularly in heavily mineralized fossils where contrast between bone and matrix is low.

For incomplete or distorted specimens, retrodeformation algorithms and bilateral mirroring (reflecting the undamaged side) can be employed to estimate the complete endocast, though these approaches introduce their own uncertainties.

While endocasts provide the best available proxy for the external morphology of the brain in extinct taxa, they have several inherent limitations. They provide no information about internal brain structure, neuronal density, connectivity, or cell composition. The space between brain and endocranial wall—occupied by meninges, vasculature, and cerebrospinal fluid—introduces systematic error, particularly in the hindbrain region. Endocasts also exhibit an artifactually 'pedomorphic' shape relative to actual brains, as demonstrated by Watanabe et al. (2019), meaning they tend to resemble younger developmental stages more than they reflect the adult brain morphology. Furthermore, ontogenetic stage can dramatically affect brain-to-endocast correspondence, making it essential to account for individual maturity when comparing across taxa.

Recent advances include the use of diffusible iodine-based contrast-enhanced CT (diceCT) to simultaneously image both the brain and endocast in the same specimen, enabling direct quantitative comparison. 3D geometric morphometrics applied to endocasts allows for sophisticated shape analyses that go beyond simple volumetric comparisons. The partitioning of endocasts into functional neuroanatomical regions—cerebrum, optic lobes, cerebellum, medulla, olfactory bulbs—enables more nuanced evolutionary analyses. Large comparative databases of digital endocasts, combined with phylogenetic comparative methods, are beginning to reveal the sequence and timing of major neuroanatomical transformations across vertebrate evolution, from the origin of the mammalian neocortex to the emergence of the avian-type brain along the theropod stem lineage.