MBExC Lecture

Artur Indzhykulian, MD, PhD, from the Massachusetts Eye and Ear, Harvard Medical School, Boston, MA will give a talk on “Harnessing AI for Enhanced Analysis of Cochlear Imaging Data” during the MBExC Lecture on 19 September, 2024 at 11:00 a.m. at the lecture hall, Max Planck Institute for Multidisciplinary Sciences, Hermann-Rein-Str. 3.

Abstract
The sensory epithelium of the mammalian cochlea exhibits a tightly organized pattern of sensory hair cells along the so-called tonotopic axis. High-resolution imaging now commonly generates large datasets from light and electron microscopy, but analyzing these massive datasets has become a bottleneck, exacerbated by the lack of efficient tools that can mitigate user biases and manual labor.
Recent advances in Artificial Intelligence and Machine Learning (AI/ML) are transforming our ability to analyze extensive datasets and accelerate scientific discovery, particularly in tasks related to bio-image analysis. We will present examples of AI/ML-based applications we have developed for analyzing large inner ear imaging datasets, demonstrating how these technologies can expedite traditional time-consuming analyses and help overcome barriers in the field. These tools serve as a blueprint for developing novel applications in the field of auditory neuroscience.
To develop one such tool, we first assembled a diverse, carefully annotated dataset comprising 2D images of auditory hair cells captured using fluorescence microscopy, contributed by the global auditory research community. We then developed an AI/ML-based application trained on this dataset that automates the detection, classification, and quantification of hair cells along the tonotopic axis. The tool leverages advanced deep learning libraries and architectures, resulting in robust, generalizable models. Next, we extended AI/ML models to a more complex challenge: analyzing serial 3D electron microscopy datasets. We developed a novel tool for volumetric instance segmentation of mitochondria, which significantly enhances the structural analysis of subcellular organelles in electron microscopy volumes.
Our results illustrate significant time savings and increased reproducibility, utilizing open-source technologies and free software to build tools that can be shared as standalone tools or ImageJ plugins. These developments streamline data processing across various imaging modalities commonly used in the field of auditory neuroscience and enable detailed, quantitative analysis of large datasets to aid in discoveries that may have been overlooked otherwise.
While not exhaustive, these case studies underscore the essential steps for developing and employing AI/ML-based tools to address complex biological questions, highlighting the potential of these technologies to advance studies that rely heavily on detailed imaging data analysis.

Host: Dr. Barbara Vona, University Medical Center Göttingen & HSC Instructor

Poster announcement: