The Imaging Revolution in Biology and Medicine

1. Why Imaging is Becoming Central

Abstract: Outside school, over both past summers, I pursued numerical math and computational biology research, where I developed regularized inverse solvers for ill-posed problems and a deep CNN for brain tumor subtype classification, respectively, at Emory College and Mayo Clinic. The commonality between the two was focused on imaging modalities, where the former involved deblurring CT scans and the latter classifying multiparametric MRI. Alongside these, I would also perform protein design & structural biology research at Winship Cancer Institute, which led me to understand the imaging modality of Cryo-EM too. Hence, this piece entails a flavor of personal interest and motivation. For the same reason, I’m lately finding myself considering my PhD interest in the intersection of engineering and computational math that I have always wanted to blend together; as I fathom, imaging comes in all spheres, from patterns in biology to signals in robotics.

Imaging as a Primary Data Modality

Imaging has turned into a primary data modality, instead of a mere diagnostic readout:

In biology: High-throughput spatial and live-cell imaging maps showcase where things are and how they behave, not just what’s there.
In structural biology: Cryo-EM, biophysical measurements, and simulation transform proteins into continuous 3D ensembles, rather than static structures.
In medicine, radiology, and pathology: Imaging is becoming pixel-level phenotypes for AI, tightly tied to genomics and clinical outcomes.
In new environments (space, microgravity): Imaging is often the only pragmatic way to observe processes as they unravel.

2. Current Landscape: Key Company Archetypes

2.1 Noetik (DCVC, etc.)

Noetik is a great example of the new generation of imaging-centric biotech companies:

Uses self-supervised ML and high-throughput spatial biology to map tumor biology and discover immunotherapy targets in the TME
Generates hundreds of terabytes of genomic, transcriptomic, and proteomic data from cells and tissues, explicitly to feed transformer-based models [1]
Built OCTO-VirtualCell, a foundation model capturing cell and tissue behavior, and Perturb-map, a high-throughput in vivo perturbation system unveiling oncological targets [2]
Raised a $40M Series A led by Polaris, with participation from Khosla, Wittington, Breakout, and DCVC [3], to advance precision oncology using these multimodal, high-dimensional datasets

2.2 Peptone (Bessemer, etc.)

Peptone arrives from an angle of structural biology and biophysical chemistry into the imaging realm:

Focuses on intrinsically disordered proteins (IDPs), highly flexible, structurally heterogeneous proteins that are often undruggable by classical structure-based methods or indeterministic by SOTA platforms, including AlphaFold
Their tech combines tailored mass spectrometry, physics-informed ML, and specialized labs to model disordered proteins and identify invisible (hidden) small-molecule pockets in the conformation [4]
Bessemer, in its Series A investment thesis, framed this as bringing order to disordered proteins via machine learning, repositioning IDPs as druggable
Raised a $40M Series A led by F-Prime and Bessemer [5] to build an advanced facility for solving complex IDP structures at scale

2.3 CZ Biohubs

Institutions like Chan Zuckerberg have pushed imaging incredibly by:

Massive investments in light-sheet microscopy [6] extending to LLSM, adaptive optics, and custom scopes
Large programs to map cellular and tissue atlases via imaging and genomics
A strong emphasis on open tools: hardware designs, software, and datasets
Creating new modalities and benchmark datasets for other methods

Similar roles are played by teams at the Allen Institute for Brain Science and Integrative Imaging at Janelia.

2.4 InkSpace Imaging

InkSpace Imaging is developing lightweight, portable MRI systems:

Utilize advanced metamaterials, flexible RF coils, low-field MRI physics, and patient-centric design
By replacing massive superconducting magnets, they enable portability
Deployable in non-hospital settings and usable for pediatric and claustrophobic patients, presents a huge market
Potentiating compatibility with bedside or emergency environments, representing an infrastructure and hardware switch

2.5 Radiology and Clinical Imaging beyond AI diagnostics

In clinical care, we’re seeing:

Radiology AI (e.g., in stroke, pulmonary embolism, oncology) turning images into structured risk scores, triage signals, and quantitative biomarkers
Pathology imaging (whole-slide scanners) is becoming more common and feeding ML for grading, subtyping, and prognosis
Functional imaging (e.g., cardiac CT/FFR, perfusion imaging) being used for computationally derived metrics and inverse problems

3. Gaps, Opportunities, and Emphasis

Areas where further potential could be tapped, not yet industrialized.

3.1 Early In-Space Imaging/Microgravity Phenotyping

Imaging of crystals, cells, tissues, and materials in low-Earth orbit [7]
Still mostly research-oriented or pilot-commercial rather than scaled business lines
Coupled tightly to biomanufacturing

3.2 Multimodal Integration is Still Nascent

Despite a variety of imaging platforms and versatile omics & clinical data, seamless fusion of imaging + genomics + transcriptomics + proteomics + EHR at the individual-patient level is nonexistent.

Key challenges [8]:

Lack of robust end-to-end tooling to go from raw images → features → multimodal joint models → decisions
Need for tools for registering and analyzing time-series imaging data across years to preserve longitudinal or dynamical modalities for disease trajectories in vivo

3.3 Imaging-Native Trial Infrastructure

Clinical trials still underuse imaging:

Imaging endpoints are often manually measured, inconsistently, and radiological & pathological images are at times siloed from biomarker and clinical data
Few clinical trials are designed with AI-readable imaging endpoints from the ground up

Opportunity: CRO/CDMO-like organizations that design and run imaging-intensive trials, roping standardized acquisition protocols, automated feature extraction, and holistic integration with molecular and clinical endpoints. This would be highly valuable in oncology, cardiology, neurology, and inflammatory diseases.

Conclusion

We see a powerful convergent thesis that whoever controls the imaging stacks, hardware, acquisition, annotations, and models would control one of the most powerful representations in bio, health, and pharma. VC firms heavily active here include: DCVC, Bessemer, a16z, General Catalyst, Khosla, Insight, etc.

References

Last updated: December 2025