Awate Research Private Limited
Proprietary MRI Evaluation Support System
Structured Anatomical Descriptors·Patent-Pending Architecture·Non-Diagnostic by Design
What we're solving
Four reasons that verdict is the wrong output. BellCNN™ is built around what should replace it.
A Verdict Is Not A Finding
Brain anatomy doesn’t reduce neatly to a yes or a no. It is continuous across regions, it changes over time, and the patterns that matter are often distributed across multiple structures at once. A binary classifier has to ignore most of that. It takes everything the scan contains and collapses it into a single output — positive or negative — because that is what the task demands of it. What gets lost in that collapse is exactly the information a clinician needs. Not a verdict. The underlying picture.
A Threshold Is Not A Standard
Every binary classifier is trained to a threshold. That threshold holds for one distribution — the one it was trained on. Change the scanner, the acquisition protocol, the field strength, or the patient population and the distribution moves. The threshold no longer reflects what it was calibrated against. The model still returns positive or negative. Nothing in that output tells you it is now operating outside the conditions it was validated on. The number looks the same. The ground has shifted.
A Label Is Not Evidence
A binary output has no audit trail behind it. No regional breakdown. No anatomical grounding. No indication of what the model weighted, or where. What gets delivered is a verdict — and a verdict, in a clinical context, needs to be traceable. It rarely is. Developers validate against their own benchmarks and ship a number. In India there is no independent framework to verify that number at scale. Going live and being safe to use are not the same thing. The difference between the two is currently not being measured.
One Bit Is Not A Clinical Instrument
A radiologist does not reason in binary. They reason across regions, across time, across degrees of certainty. A system that returns a single label removes the structure that makes that kind of reasoning possible. What is needed at the point of clinical decision is not a verdict — it is a set of structured anatomical descriptors that a qualified expert can interrogate. No existing system delivers that within the imaging workflow, without a separate pipeline, without collapsing the output into a forced binary conclusion.
Architecture
When a brain MRI is processed by BellCNN™, it travels through two parallel pathways at once. One works across 2D slice projections, picking up local anatomical patterns at the level of individual planes. The other processes the full 3D volume, preserving the spatial relationships that only become visible when you look at the brain as a whole. These two representations converge at a fusion layer — and what comes out the other side is not a verdict. It is a set of structured anatomical descriptors, each one anchored to a named neuroanatomical region.
The system-level architecture is covered under Patent App. 202621005875. The output is intended for qualified researchers and domain experts — not for unsupervised clinical use.
The descriptors BellCNN™ produces are not free-floating learned features. They are grounded in an atlas of anatomical regions — which means each output corresponds to a part of the brain that has a name, a location, and an established clinical reference. That grounding is what makes the output auditable. It is also what makes it useful beyond a single task or dataset.
The system also monitors the quality of its own spatial registration. When a scan doesn’t meet the threshold, it flags it — clearly, in the output record. It doesn’t silently discard the data or return a result that looks valid when it isn’t. BellCNN™ is registered as a Class 10 Medical Apparatus under the Trade Marks Act, 1999.
BellCNN™ produces assistive analytical outputs for qualified medical professionals only. No automated clinical decisions are produced or implied.
01
Structured Descriptor Output
Three categories of anatomical descriptors — regional morphometric metrics, latent neural embeddings, and volumetric indicators — not binary diagnostic labels.
02
Dual-Path Architecture
Parallel 2D slice and 3D volumetric branches, fused at inference. Local and global spatial context captured simultaneously.
03
Anatomically Anchored
Outputs are grounded in recognised neuroanatomical regions. Descriptors correspond to named brain structures, not arbitrary learned features.
04
In-Situ Processing
PACS integration in progress. Designed to operate within the imaging workflow — no manual data export, no separate inference pipeline.
Services
Awate Research offers the following services to support its R&D efforts. The practice is grounded in active peer review — with no affiliation to any AI developer, vendor, or system being assessed. Available to academic institutions, research hospitals, and medical imaging AI developers.
Academic & Institutional
Independent Assessment
A structured, written review of a medical imaging AI system — examining methodological soundness, reproducibility across datasets, scanner and demographic variability, and statistical integrity. The assessment is vendor-independent. It is delivered as a document, not a score.
Researchers & Authors
Manuscript Review
A pre-submission review from the perspective of someone who has sat on the other side of the process 54 times. Covers methodology, statistical approach, presentation clarity, and how the paper is likely to read to a reviewer. Designed to find the problems before the journal does.
All Tiers
Institutional Advisory
For institutions that are building, acquiring, or deploying medical imaging AI — and want an independent perspective on how to do it properly. Covers evaluation framework design, validation protocol structure, and methodology review. Available on a retainer or engagement basis.
Research & Supervised Pilot · Hospitals & OEM
BellCNN™ Evaluation Support
Structured MRI evaluation using BellCNN™ in formal research and institutional validation contexts. This is not a general-access service. It requires an institutional engagement and is conducted under qualified supervision. Not available for unsupervised clinical use.
All engagements begin with a no-obligation scoping call · General enquiry →
Roadmap
OEM integration is the architectural endpoint — the horizon the system is designed for from the ground up.
Where We’re At
Funding applications active
The PACS integration build and a structured hospital pilot. This phase is contingent on securing further funding. If funded, BellCNN™ moves from a supervised research tool to a deployable clinical evaluation product. That transition is also what makes OEM conversations credible.
Our Eventual Goal
Post-pilot
The endpoint the architecture has been pointed at from the beginning. BellCNN™ embedded directly into imaging hardware and clinical software pipelines — structured descriptor generation running natively within the scanner environment, without a separate export step, without a separate pipeline. OEM partnerships. Product-level recurring revenue. And the foundation for the next generation of research tracks: multimodal extension, longitudinal monitoring, normative modelling.
Founder
Gururaj Awate
Founder · Awate Research Private Limited
Background
Gururaj Awate has been reviewing medical imaging AI research for IEEE and Elsevier for years. Fifty-four verified reviews across IEEE Access, Neural Networks, and Pattern Recognition Letters — and a Certified Publons Academy Mentor credential that puts him at the 99th percentile globally. That practice is not background context. It is the professional foundation the entire service offering is built on.
Why It’s Built This Way
The design philosophy of BellCNN™ — its insistence on transparency, auditability, and keeping qualified humans in every decision loop — did not emerge from a literature review. It comes from direct experience as a caregiver navigating clinical decisions where the gap between a system’s output and a clinician’s accountability was never clearly defined. That experience shaped what the system is, and what it deliberately is not.
Parallel Research Interest
Separately from BellCNN™, Gururaj has a long-running research interest in assistive technology — specifically BCI systems for hemiplegia, bilateral motor mirroring, and the role of mirror neuron activation in motor rehabilitation. It is hypothesis-stage. It is long-horizon. It is not on the company roadmap. But it comes from the same instinct that shaped BellCNN™ — the idea that the most important thing a system can do is work for the person in front of it.
Team & Collaborators
Awate Research works with domain experts — radiologists, clinicians, PhD researchers, and medical professionals — as contracted collaborators. Technical and engineering work is carried out by contracted personnel.
Demonstrations & Materials
V2026 access is restricted to current institutional partners. Prospective collaborators may request access via the contact form below.
BellCNN™ Demo — 2019 Architecture, Updated for 2026 (V2019_Mod2026)
Contact
For research collaborations, service enquiries, institutional engagement, or anything else — reach out directly. All communications are handled by Awate Research.