**Author Listing - Lead/ Submitting Author (we will be communicating with this individual): Name, Institution, City and State where the work was done. Co-authors will be added in a separate section.**
Taxiarchis Kourelis, Mayo Clinic, Rochester, MN

**Main Body of the Abstract**
Background: Cardiac AL/ATTR amyloidosis is a potentially fatal cardiomyopathy. No well accepted methods for amyloid quantification exist. It is unclear how well imaging studies (Scintigraphy, MRI) correlate with histologic amyloid burden. Traditional histologic scoring systems can be biased, low throughput, semi-quantitative and difficult to generalize. This study aims to quantify amyloid burden in endomyocardial biopsies by training a computer vision model to identify endocardial, vascular, fat and interstitial deposits. Methods: Whole slide images of SAB stains from 15 unique specimens captured on an Aperio GT 450 scanner were used for model training in Aiforia Cloud Create. Trained experts annotated 384 regions in the training images and the model was trained for 1000 iterations using high complexity, a 50um field of view and minimal image augmentation. Other parameters were adjusted in parallel to train 8 slightly different models, in time for submission to this meeting. Validation was performed in a distinct hold-out set of seven images comprising 56 regions. Results: Initial results in this limited validation set and considering the best performing model demonstrated a total error area error rate of 8.1%, an F1 score of 71.57% a precision of 74%, sensitivity if 69%. Main sources of error include images with different staining intensity and misclassification of amyloid location (vessels, fat etc). Conclusions: These early results suggest that artificial-intelligence based amyloid quantification in cardiac biopsies is feasible. After additional refinement, this approach will allow us to correlate histologically defined burden with outcomes, radiologic and laboratory parameters.



**Author Listing - Lead/ Submitting Author (we will be communicating with this individual): Name, Institution, City and State where the work was done. Co-authors will be added in a separate section.**
Taxiarchis Kourelis, Mayo Clinic, Rochester, MN

**Main Body of the Abstract**
**Background**: Cardiac AL/ATTR amyloidosis is a potentially fatal cardiomyopathy. No well accepted methods for amyloid quantification exist. It is unclear how well imaging studies (Scintigraphy, MRI) correlate with histologic amyloid burden. Traditional histologic scoring systems can be biased, low throughput, semi-quantitative and difficult to generalize. This study aims to quantify amyloid burden in endomyocardial biopsies by training a computer vision model to identify endocardial, vascular, fat and interstitial deposits. **Methods**: Whole slide images of SAB stains from 15 unique specimens captured on an Aperio GT 450 scanner were used for model training in Aiforia Cloud Create. Trained experts annotated 384 regions in the training images and the model was trained for 1000 iterations using high complexity, a 50um field of view and minimal image augmentation. Other parameters were adjusted in parallel to train 8 slightly different models, in time for submission to this meeting. Validation was performed in a distinct hold-out set of seven images comprising 56 regions. **Results**: Initial results in this limited validation set and considering the best performing model demonstrated a total error area error rate of 8.1%, an F1 score of 71.57% a precision of 74%, sensitivity if 69%. Main sources of error include images with different staining intensity and misclassification of amyloid location (vessels, fat etc). **Conclusions**: These early results suggest that artificial-intelligence based amyloid quantification in cardiac biopsies is feasible. After additional refinement, this approach will allow us to correlate histologically defined burden with outcomes, radiologic and laboratory parameters.



A feasibility study into computer vision quantification of cardiac amyloid burden in endomycocardial biopsies OA1 (#496)

We invite you to attend the XIX International Symposium on Amyloidosis to be held in Rochester, MN.

This symposium is organized as a hybrid meeting (with the focus on in-person) held from Sunday, May 26 through Thursday, May 30, 2024.

Together with the ISA Board and members of Mayo Clinic, a very attractive scientific hybrid program is being planned. The program will cover a variety of types of systemic amyloidosis: ATTR (hereditary and wild-type), AL (light-chain), AA and less frequent forms. There will be key lectures, specific symposia including those company sponsored, oral presentations and poster sessions of accepted abstracts.

COURSE DESCRIPTION
This program will provide a comprehensive review of state-of-the-art education in systemic amyloidosis: basic science, pathophysiology, diagnosis, current and emerging treatments. Scientific presentations will include competitive oral and poster abstracts. The rapid evolution of scientific discovery and expanding treatment options has created tremendous interest in the field and this meeting will fill and important gap in educating scientists and clinicians.