IBM Watson Health Liver Advisor — medical imaging AI

IBM Watson Health — Liver Advisor (CT & MRI)

Three production releases of liver and lesion detection in CT and MRI, validated against expert radiologists in pilot studies. Co-inventor on the lesion-detection patents listed below.

Status: completed · 2018 – 2022

Role: Advisory Data Scientist on the Liver Advisor program at IBM Watson Health (the enterprise imaging portfolio later marketed under Merative). The Liver Advisor team grew to roughly thirteen engineers and scientists at peak, with many more patents across the program; the patents listed below are the ones Omid is named on as co-inventor. Alongside that day-to-day engineering, IBM Watson Health publicly partnered with Guerbet—a French imaging company—from 2018; that storyline ended under Merative in 2022 with co-developed assets assigned back to Guerbet. Patents and publications below are verifiable independently; the sections that follow summarize only what appears in those public disclosures.

Strategic arc: Guerbet, IBM Watson Health, and Merative

In July 2018, Guerbet—a global diagnostic and interventional imaging company headquartered in France—and IBM Watson Health announced an exclusive joint development agreement centered on liver cancer diagnostics using CT and MRI. Press materials describe a planned first deliverable working under names such as Watson Imaging Care Advisor for Liver (a working label, not necessarily a finalized commercial or regulatory product name at the time). This relationship is distinct from hospital ownership: Guerbet is an imaging vendor, not a single-site liver hospital.

Subsequently, IBM reorganized and sold its healthcare data and analytics portfolio; Watson Health Imaging became part of what is now marketed as Merative Imaging. Institutional branding changed; the Liver Advisor engineering lineage continued under that umbrella.

On 3 November 2022, Guerbet publicly announced an agreement with Merative that formalized the end of their collaboration—which Guerbet dates to July 2018—and stated that an asset-assignment arrangement would transfer co-developed outputs to Guerbet, including application source code, AI algorithms, and related intellectual property, so Guerbet could steer its own roadmap. Merative framing at the time pointed to a strategic shift in how it packaged AI inside its broader imaging offerings.

Later industry news—for example announcements through 2024 involving Guerbet and other imaging-software partners on liver lesion AI and regulatory milestones—suggests sustained commercial interest in the problem space. None of those later items substitute for firsthand knowledge of what any given hospital deployed and when.

Published sources (external links)

Primary references for the partnership and exit narrative above.

Closing the loop honestly

This case study is written from the perspective of TensorHarmony's principal, who contributed to Liver Advisor-era engineering at IBM. It does not cite confidential customer lists, unpublished deployment data, or private Merative / Guerbet roadmap details.

Public announcements prove that a Guerbet–IBM liver-AI program existed, that IBM’s imaging business passed through Merative, and that Guerbet later received co-developed assets—but they do not, by themselves, tell us whether a specific institution today is running Liver Advisor-era algorithms under the original branding, nor the extent of routine clinical use. Readers should treat later Guerbet or partner regulatory milestones as evidence of ongoing productization efforts, not as proof that any particular deployment matches the IBM production releases summarized in the outcomes below unless independently verified.

Scope of work

Day-to-day responsibilities centered on machine learning and deep learning for liver CT and MRI analysis, in service of three production releases (two CT, one MRI). The work spanned modeling, data quality, generative augmentation, pipeline engineering, and mentorship.

Liver and lesion segmentation models (Dense-U-Net family) on 3D, multi-parametric CT and MRI.
Liver boundary estimation: LSTM-based prediction with StyleGAN-style mapping for structured conditioning.
Anomaly detection for rejecting non-liver MRI scans (boundary features + SVM).
MRI phase and sequence classification (DeepLSTM with mapping-network conditioning).
Generative R&D: GANs for lesion synthesis, denoising, and domain adaptation.
CycleGAN-based synthetic lesion embedding in normal liver CT.
Noise2Noise-style MRI restoration without clean targets.
Low-memory model customizations using pre-trained ImageNet backbones; novel loss functions and sampling strategies.
Semi-automated data curation, annotation, and quality-check pipeline used heavily by the liver team.
Production pipeline engineering in C++/Python: bug fixes, multithreading, scalability.

Outcomes

3

Production releases (2 CT, 1 MRI)

2

Co-invented lesion-detection patents

2

Peer-reviewed publications (RSNA, IEEE ISBI)

Granted US patents (co-inventor — lesion detection)

Ensemble machine learning model architecture for lesion detection

US 11688063 B2

27 June 2023

Lesion detection artificial intelligence pipeline computing system

US 11436724 B2

6 September 2022

Selected publications

Evaluation of an Automated Method to Detect Missed Focal Liver Findings in Single-Phase CT Images of the Abdomen

2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI) · 26 April 2022

Recognition

IBM People's Choice Culture Award (Nov 2020) — recognized for contributions to the IBM Watson Health Imaging team, including AI algorithms for detection and classification of liver lesions in CT.

Discuss Your Next System or Deployment

Technical discovery, production engineering, and deployment-focused collaboration for intelligent systems, computer vision, and medical imaging workflows.

Schedule a Technical Discussion

No obligation — just a focused technical conversation