Hearing loss is an important public healthcare problem that affects the quality of life for 1.5 Billion people around the Globe. Those suffering from hearing loss have an increased risk of dementia. These problems are expected to worsen with the globally aging population. It is not surprising, therefore, that improved access to hearing aids is one of the key actions recommended by the World Health Organization (WHO, 2018). The morphology of the outer ear canal is not only highly important for a good hearing aid fit (Paulsen et al., 2002), but also essential for a wide range of scientific and clinical purposes (Voss et al., 2020). There has been, however, limited data available that can be used to understand the ear’s anatomy and variations among individuals (Voss et al., 2020). Fortunately, the advent of technologies such as 3D scanning, computer-aided design and additive manufacturing transformed the design, production and fitting of hearing aids (Valente et al., 1998) simplifying the measurement of the ear canal’s cavity (Pirzanski, 2006), integrating electrical and mechanical components, and 3D printing, rather than casting, bespoke in-the-ear shells (Sickel et al., 2011, Baloch et al., 2010). Nevertheless, creating a digital model still requires injecting silicone into the ear canal to produce an Ear Canal Impression (ECI) which is afterward digitized using a 3D scanner (Sullivan, 2007). Today it is common for audiologists to use 3D scanners to create digital models of ECIs and large volumes of human ear canal scans now exist. This offers an opportunity to study the ear canal’s morphology using computational methods. Indicatively, 10,000 ECIs were available here, whereas in past works only a few dozen (Stinson and Lawton, 1989, Paulsen et al., 2004) to low hundreds were examined (Baloch et al., 2010, Voss et al., 2020). However, before the data can be useful they must be substantially 3D edited. Indicatively, only 3000 ECIs were used due to the manual labor required to pre-process and normalize the data. Pre-processing ECIs presents two major challenges that stem from (a) the ambiguity of associating anatomical semantics with geometric features, and (b) the discrepancies introduced by creating and digitizing the silicone impressions.

While the anatomy of the outer ear has been extensively studied in the past (Alvord and Farmer, 1997, Møller, 2012) and regions of the ear’s surface have been qualitatively attributed, translating those descriptions into quantifiable geometric features is non-trivial (Baloch et al., 2010). Additionally, inconsistencies in the silicone injection process, and modifications to the ear molds or 3D scans thereof, introduce substantial variations among ECIs. For instance, the scanned ear canal’s length is affected by the insertion depth of a cotton block, which is used to protect the patient’s eardrum, before injecting the silicone. Moreover, the boundary of the external ear surface is also determined by the volume of silicone used. Excessive silicone is often removed from ECIs by the clinician with a knife, creating surfaces irrelevant to the ear’s morphology. It is thus difficult to unambiguously discern the parts of the 3D geometry that were in contact with the ear canal’s surface. Finally, 3D scanners use sensor-oriented coordinate systems and the placement of ECIs within these devices is often inconsistent. The objective of this work is to overcome the challenge of ECI data pre-processing aiming to assist hearing aid designers, by offering tools that can reduce manual geometry 3D editing; audiologists, by providing coherent and conveniently visualized representations of the ear canal and data scientists, by offering tools for normalizing large datasets.