Freya Schulze
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Freya Schulze has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Jan-Niklas Eckardt
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
3 shared publications - Susann Winter
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
3 shared publications - Jan Moritz Middeke
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
2 shared publications - Katja Sockel
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
2 shared publications - Martin Bornhäuser
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
2 shared publications - Mohamed Benouis
Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Dresden, Germany
2 shared publications
Latest Publications
- Open Access
Multimodal Artificial Intelligence Fusion Models to Predict Response to Venetoclax and Azacitidine in Patients with Newly Diagnosed Acute Myeloid Leukemia
Venetoclax and azacitidine (ven-aza) is a standard of care therapy for newly diagnosed acute myeloid leukemia (AML) patients unfit to receive intensive therapy, and may also be appropriate for certain fit patients. Models predicting response to ven-aza currently focus predominantly on genetics. Here, we present two multimodal fusion models: a custom ResNet-34-based model and a pathology foundation model fine-tuned for bone marrow smear (BMS) images that combine clinical and genetic variables with whole slide images. Training on 193 newly diagnosed AML patients who received ven-aza, we obtained an area-under-the-receiver-operating-characteristic (AUROC) of 0.79 and 0.82 for the ResNet-based and fine-tuned foundation models in a heldout test set, and 0.76 and 0.73 in an external validation set, predicting best response to this therapy. Sensitivity analysis showed that fusing image data with clinical and genetic variables outperformed training on either modality alone. In univariable and multivariable analyses, image-based classification was an independent predictive marker, second only in effect size to IDH1/2 mutation status, outperforming other genetic alterations. Image-based explainability showed high model attention on myeloblasts, suggesting previously unrecognized cell-specific patterns associated with response. Our findings suggest BMS images complement genetic biomarkers to refine outcome prediction in ven-aza treatment and incentivize AI-supported outcome prediction, including bone marrow morphology.
Open AccessMultimodal Artificial Intelligence Fusion Models to Predict Response to Venetoclax and Azacitidine in Patients with Newly Diagnosed Acute Myeloid Leukemia
Venetoclax and azacitidine (ven-aza) is a standard of care therapy for newly diagnosed acute myeloid leukemia (AML) patients unfit to receive intensive therapy, and may also be appropriate for certain fit patients. Models predicting response to ven-aza currently focus predominantly on genetics. Here, we present two multimodal fusion models: a custom ResNet-34-based model and a pathology foundation model fine-tuned for bone marrow smear (BMS) images that combine clinical and genetic variables with whole slide images. Training on 193 newly diagnosed AML patients who received ven-aza, we obtained an area-under-the-receiver-operating-characteristic (AUROC) of 0.79 and 0.82 for the ResNet-based and fine-tuned foundation models in a heldout test set, and 0.76 and 0.73 in an external validation set, predicting best response to this therapy. Sensitivity analysis showed that fusing image data with clinical and genetic variables outperformed training on either modality alone. In univariable and multivariable analyses, image-based classification was an independent predictive marker, second only in effect size to IDH1/2 mutation status, outperforming other genetic alterations. Image-based explainability showed high model attention on myeloblasts, suggesting previously unrecognized cell-specific patterns associated with response. Our findings suggest BMS images complement genetic biomarkers to refine outcome prediction in ven-aza treatment and incentivize AI-supported outcome prediction, including bone marrow morphology.
- Open Access
Benchmarking ten frontier large language models on 1,477 board style multiple choice questions in hematology
Large Language Models (LLMs) are increasingly used by clinicians and patients for medical queries, yet their accuracy and safety at the specialist level in hematology remain insufficiently characterised. We benchmarked ten frontier proprietary and open-weight LLMs across two generations on 1,477 board-style hematology multiple-choice questions (MCQs) derived from five educational datasets spanning nine disease areas and six clinical skill domains, including text-only and multimodal case vignettes. Claude Opus 5 had the highest mean accuracy (92.7% text, 76.9% multimodal), followed closely by Gemini-3.1 Pro (91.4% and 78.7%), Gemini-3.6 Flash (91.0% and 74.8%) and GPT-5.6 Sol (89.9% and 76.7%). Accuracy significantly correlated with model size both for text-only and multimodal MCQs. Between model generations, the largest improvements in accuracy were seen for open-weight models whereas proprietary models showed only marginal gains. In error analysis, top-performing models exhibited highly concordant failure patterns, suggesting shared limitations on challenging cases. Frontier LLMs exhibit substantial specialist hematology knowledge across diverse subspecialist domains and clinical skill sets. Yet, despite high accuracy on board-style questions in hematology, continuous expert-on-the-loop output monitoring is paramount.
Open AccessBenchmarking ten frontier large language models on 1,477 board style multiple choice questions in hematology
Large Language Models (LLMs) are increasingly used by clinicians and patients for medical queries, yet their accuracy and safety at the specialist level in hematology remain insufficiently characterised. We benchmarked ten frontier proprietary and open-weight LLMs across two generations on 1,477 board-style hematology multiple-choice questions (MCQs) derived from five educational datasets spanning nine disease areas and six clinical skill domains, including text-only and multimodal case vignettes. Claude Opus 5 had the highest mean accuracy (92.7% text, 76.9% multimodal), followed closely by Gemini-3.1 Pro (91.4% and 78.7%), Gemini-3.6 Flash (91.0% and 74.8%) and GPT-5.6 Sol (89.9% and 76.7%). Accuracy significantly correlated with model size both for text-only and multimodal MCQs. Between model generations, the largest improvements in accuracy were seen for open-weight models whereas proprietary models showed only marginal gains. In error analysis, top-performing models exhibited highly concordant failure patterns, suggesting shared limitations on challenging cases. Frontier LLMs exhibit substantial specialist hematology knowledge across diverse subspecialist domains and clinical skill sets. Yet, despite high accuracy on board-style questions in hematology, continuous expert-on-the-loop output monitoring is paramount.
- Open Access
Vision Language Models Fail to Reliably Detect Acute Myeloid Leukemia in Bone Marrow Smears
Hematologic diagnostics and especially cytomorphologic assessment are time-intensive and require high levels of expertise. Vision Language Models (VLM) show promise in medical image analysis in radiology and histopathology, while an evaluation on detecting acute myeloid leukemia (AML) is lacking. Our goal was to evaluate three Vision Language Models regarding their diagnostic accuracy and safety in clinical decision support in detecting AML from digitized bone marrow smears (BMS). Whole slide images were obtained from bone marrow smears of 50 AML patients and 50 bone marrow donors. Ten representative fields of view per sample were extracted manually. Three VLMs were used, two of which are considered generalist models (Qwen3.5-397B-A17B-FP8, GLM-4.6V-FP8), while the other one is a medically adapted model (Medgemma-27b-it). All models performed zero-shot analysis using two prompting strategies: First, a context-rich prompt requesting reporting of WHO/FAB diagnostic criteria in a structured manner, and secondly a minimal prompt without specific hematologic context. Overall diagnostic accuracy was poor for all models as they exhibited the overwhelming tendency to classify most samples as leukemic: With context-rich prompts, GLM4.6 identified 90% of leukemic samples while also labeling 92% of bone marrow donors as AML. The medical specialist model MedGemma-27b showed similar failure, misclassifying 86% of healthy donors and correctly detecting AML in only 66% of cases. Qwen3.5 performed best under detailed prompting, achieving a specificity of 0.26 and accuracy of 0.51. Accuracy of all models improved with context-free prompts (accuracies range 0.47-0.79), yet they still lacked the ability to correctly distinguish between leukemia and healthy bone marrow. Qwen3.5 was the only model to maintain meaningful specificity (0.64) and correctly identified 94% of AML, yielding an overall accuracy of 0.79. Morphologic feature-level agreement with human expert reports was poor across all models, indicating poor recognition of cell-level morphologies. This failure is likely driven by the fact that pathology imaging archives are vastly scraped during model training while hematological samples are not as widely available and therefore, hematology is an out-of-bounds use-case for these models, rendering them currently unsuitable for clinical decision support in hematology.
Open AccessVision Language Models Fail to Reliably Detect Acute Myeloid Leukemia in Bone Marrow Smears
Hematologic diagnostics and especially cytomorphologic assessment are time-intensive and require high levels of expertise. Vision Language Models (VLM) show promise in medical image analysis in radiology and histopathology, while an evaluation on detecting acute myeloid leukemia (AML) is lacking. Our goal was to evaluate three Vision Language Models regarding their diagnostic accuracy and safety in clinical decision support in detecting AML from digitized bone marrow smears (BMS). Whole slide images were obtained from bone marrow smears of 50 AML patients and 50 bone marrow donors. Ten representative fields of view per sample were extracted manually. Three VLMs were used, two of which are considered generalist models (Qwen3.5-397B-A17B-FP8, GLM-4.6V-FP8), while the other one is a medically adapted model (Medgemma-27b-it). All models performed zero-shot analysis using two prompting strategies: First, a context-rich prompt requesting reporting of WHO/FAB diagnostic criteria in a structured manner, and secondly a minimal prompt without specific hematologic context. Overall diagnostic accuracy was poor for all models as they exhibited the overwhelming tendency to classify most samples as leukemic: With context-rich prompts, GLM4.6 identified 90% of leukemic samples while also labeling 92% of bone marrow donors as AML. The medical specialist model MedGemma-27b showed similar failure, misclassifying 86% of healthy donors and correctly detecting AML in only 66% of cases. Qwen3.5 performed best under detailed prompting, achieving a specificity of 0.26 and accuracy of 0.51. Accuracy of all models improved with context-free prompts (accuracies range 0.47-0.79), yet they still lacked the ability to correctly distinguish between leukemia and healthy bone marrow. Qwen3.5 was the only model to maintain meaningful specificity (0.64) and correctly identified 94% of AML, yielding an overall accuracy of 0.79. Morphologic feature-level agreement with human expert reports was poor across all models, indicating poor recognition of cell-level morphologies. This failure is likely driven by the fact that pathology imaging archives are vastly scraped during model training while hematological samples are not as widely available and therefore, hematology is an out-of-bounds use-case for these models, rendering them currently unsuitable for clinical decision support in hematology.