Simplex Research

Supporting Research and Innovation in Healthcare with Artificial Intelligence

We support AI-based radiology and pathology research projects and carry out research and development across a range of fields, starting from clinical needs. Our goal is to turn artificial intelligence into reliable, practical solutions with high clinical value in healthcare.

To this end, we build R&D-focused collaborations with industry stakeholders, healthcare institutions, universities and academic research groups, working together on projects that address real clinical needs. By bringing together data, clinical expertise, artificial intelligence and software engineering, we help research ideas grow from prototypes into practical healthcare technologies.

With a focus on radiology and digital pathology, we run projects in image analysis, AI-assisted decision support systems, risk classification and clinical research, aiming to build a sustainable R&D ecosystem that strengthens collaboration between academia and industry.

We collaborate with academic and clinical partners to develop AI-driven solutions that advance research, improve clinical workflows, and shape the future of healthcare.

Research Areas

Digital Pathology and Radiology Research

We carry out studies on imaging technologies, digital workflows and clinical data management in digital pathology and radiology. We develop innovative technologies that support research processes and contribute to the digital transformation of healthcare.

AI Research in Pathology and Radiology

We develop artificial intelligence solutions in medical image analysis, automated assessment, risk classification and clinical decision support systems. With algorithms built around clinical needs and data-driven approaches, we support the efficiency and standardization of diagnostic processes.

R&D and Innovation in Pathology and Radiology

In collaboration with academics, healthcare professionals and research institutions, we turn clinical needs into innovative technology solutions. Through interdisciplinary R&D projects, we support scientific research and contribute to the development of next-generation healthcare technologies.

HEMATOPATHOLOGY · RESEARCH

Artificial Intelligence-Based Segmentation of Megakaryocytes in Bone Marrow Biopsies

Telling the subtypes of myeloproliferative neoplasms apart is a demanding task in routine pathology. The Department of Pathology at Ankara University Faculty of Medicine and the Simplex Artificial Intelligence Team developed models that detect and classify megakaryocytes in bone marrow biopsies with artificial intelligence and carry this information into diagnosis.

The clinical problem: separating three look-alike diseases

Polycythemia vera (PV), essential thrombocythemia (ET) and primary myelofibrosis (PMF) are the three classical subtypes of BCR::ABL1-negative myeloproliferative neoplasms (MPNs). Although they share overlapping clinicopathological and molecular features, they differ in complications and disease progression, which makes them hard to distinguish in routine practice.

Artificial intelligence can strengthen conventional morphological assessment through quantitative image analysis. The study aimed to develop machine learning models that help pathologists diagnose MPNs and differentiate their subtypes by identifying and analyzing megakaryocyte features.

Method: digital slides, expert labels, AI

Hematoxylin and eosin-stained bone marrow biopsies from 27 PV, 37 ET and 17 prefibrotic PMF cases diagnosed between January 2012 and September 2024, together with 26 non-neoplastic biopsies from patients without hematolymphoid disease or prior chemotherapy, were reviewed retrospectively and digitized with a 3DHISTECH scanner.

In the digital images, megakaryocyte nuclei (normal, bare, separated, bulbous, hyperchromatic, hyperlobated, and monolobated/hypolobated/micromegakaryocytes), cytoplasmic contours and clusters were annotated manually in the Simplex PACS software. Using these labeled data, AI algorithms were developed for megakaryocyte segmentation, classification and MPN subtype differentiation based on cytomorphologic and topographic features.

Results

The segmentation model detected megakaryocytes with an F1 score of 87.6%. The classification model distinguished dysplastic from normal megakaryocytes with 96.74% accuracy and predicted nuclear morphologic classes with 92.53% accuracy.

Diagnostic models differentiated MPNs from non-neoplastic bone marrow with an average accuracy of 94%, reaching 100% in the best-performing model. For subtype differentiation, accuracy was 81.48% with morphological features alone and rose to 92.59% once clinical and laboratory data were added.

Conclusion

AI-based models showed strong performance in megakaryocyte segmentation and classification and in the diagnosis and subtype differentiation of MPNs. These approaches may support pathologists by automating workflows, accelerating diagnostic processes and providing objective results. The findings point to potential clinical applicability and adaptability to other hematologic malignancies that require quantitative histologic evaluation.

COLLABORATIONDepartment of Pathology, Ankara University Faculty of Medicine
ARTIFICIAL INTELLIGENCESimplex Artificial Intelligence Team
FRAMEWORK107 bone marrow biopsies (27 PV, 37 ET, 17 prefibrotic PMF, 26 non-neoplastic) + annotation in Simplex PACS
OUTCOMESpecialty thesis
PATHOLOGY · RESEARCH

Using Artificial Intelligence for Risk Stratification in Pheochromocytomas and Paragangliomas

Predicting which pheochromocytomas and paragangliomas will metastasize is one of the hardest questions in pathology. The Department of Pathology at Ankara University Faculty of Medicine and the Simplex Artificial Intelligence Team set out to explore how artificial intelligence can help answer it.

The clinical problem: predicting metastatic risk

Pheochromocytomas and paragangliomas are neuroendocrine tumors arising from the adrenal medulla and the extra-adrenal paraganglia. Estimating their metastatic risk in advance is essential for determining a patient’s prognosis and planning the right follow-up strategy, yet it remains a significant challenge in clinical practice.

The approach: combining GAPP with artificial intelligence

In the study, tumors were evaluated histopathologically according to the GAPP (Grading system for adrenal phaeochromocytoma and paraganglioma) framework. These findings were then combined with artificial intelligence methods to investigate whether computational models can make risk stratification more effective.

Where clinical expertise meets engineering

The research also formed the core of a specialty thesis at Ankara University. It brought together the pathologists’ expertise in histopathological assessment and Simplex’s experience in artificial intelligence and software engineering within a single project.

Simplex’s approach to R&D

We believe meaningful innovation in medical AI is only possible through close collaboration between clinicians, researchers and technology teams. That is why we continue to invest our engineering expertise and AI resources in academic research, particularly in pathology and medical imaging. Our aim is to turn scientific ideas into technologies that can, over time, contribute to clinical practice.

COLLABORATIONDepartment of Pathology, Ankara University Faculty of Medicine
ARTIFICIAL INTELLIGENCESimplex Artificial Intelligence Team
FRAMEWORKGAPP grading system + AI methods
OUTCOMESpecialty thesis at Ankara University
RADIOLOGY · PROOF OF CONCEPT

Detecting Vertebral Change in Follow-up CT with Radiomix

Can texture tell us a vertebra has changed? In this proof of concept, every vertebra was segmented automatically, 172 radiomics features were extracted from each, and every follow-up CT was compared with the first one, to point the radiologist to the vertebra that changed differently from the rest of the spine.

Why relative to the rest of the spine?

Following the mean HU of the trabecular core showed that most vertebrae lose density together; the whole-spine decline was about 20%. That is why change was measured relative to the rest of the spine rather than in absolute terms: the typical (median) change of the other vertebrae was subtracted from a vertebra’s change and the result divided by the normal spread.

Method: from DICOM to a ranked list of vertebrae

The pipeline is fully automatic. On 7 non-contrast CT scans acquired with the same protocol (GE, 120 kVp, 5 mm), SPINEPS outlined and named every vertebra (T7–L5). The region of interest was the vertebral body (cortex and endplates included); partly imaged vertebrae were excluded.

Radiomix extracted 172 features from each region across intensity, texture, shape and IBSI groups. Each scan was compared both with the baseline and with the previous scan. Vertebrae that changed differently from the rest of the spine were ranked for review.

Findings

Vertebra naming stayed consistent across all 7 scans (volumes and positions stable). The whole-spine decline in density could be separated from change specific to a single vertebra.

The strongest, persistent signal was in L1. About two months after the first scan local dissimilarity (Geary’s C) rose, and about four months after it kurtosis, skewness and maximum HU rose: a few very bright voxels in an otherwise uniform body. This matches a small bright focus visible in L1 from the same period and needs radiologist review.

T9 was denser than the others from about the second to the sixth month after the first scan; its baseline value was also atypical. This is not an edge effect.

Limitations and next steps

There is no normal range yet: the scores rank, they do not decide “abnormal”. The study rests on one patient, 5 mm slices and about 8 vertebrae per scan to estimate the spread. There is no calibration phantom or reference tissue, and no radiologist ground truth yet.

Next steps: radiologist reading of L1 and T9, defining normal change on patients without new lesions, thin-slice reconstructions when available, and testing on cases with known lesion development (Spine-Mets cohort).

DATAOne patient, 7 non-contrast CT scans over about a year (same protocol, 5 mm); 11 vertebrae (T7–L5)
METHODSPINEPS segmentation + Radiomix (172 features per region) + change score relative to the spine
OUTPUTA ranked list of vertebrae for review; no diagnosis is made, the output is “look here”
STATUSProof of concept · research use only, not for clinical use

Shaping the Future of Healthcare Technology Through Research and Collaboration