A5055957832- COVID-19 diagnosis using AI
- AI in cancer detection
- Cell Image Analysis Techniques
- Image Processing Techniques and Applications
- Domain Adaptation and Few-Shot Learning
- Radiomics and Machine Learning in Medical Imaging
- Computational Drug Discovery Methods
- Sentiment Analysis and Opinion Mining
- Anomaly Detection Techniques and Applications
- Modular Robots and Swarm Intelligence
- Machine Learning and Data Classification
- Brain Tumor Detection and Classification
- Biosimilars and Bioanalytical Methods
- Advanced Neural Network Applications
- Reinforcement Learning in Robotics
- Data Management and Algorithms
- Machine Learning and Algorithms
- Complex Network Analysis Techniques
- Robot Manipulation and Learning
- Artificial Intelligence in Healthcare and Education
- Image Retrieval and Classification Techniques
- Artificial Intelligence in Healthcare
- Digital Imaging for Blood Diseases
- Single-cell and spatial transcriptomics
- Topic Modeling
KTH Royal Institute of Technology
2020-2024
Science for Life Laboratory
2020-2024
AstraZeneca (Sweden)
2022-2024
Convolutional Neural Networks (CNNs) have reigned for a decade as the de facto approach to automated medical image diagnosis. Recently, vision transformers (ViTs) appeared competitive alternative CNNs, yielding similar levels of performance while possessing several interesting properties that could prove beneficial imaging tasks. In this work, we explore whether it is time move transformer-based models or if should keep working with CNNs - can trivially switch transformers? If so, what are...
Transfer learning is a standard technique to transfer knowledge from one domain another. For applications in medical imaging, ImageNet has become the de-facto approach, despite differences tasks and im-age characteristics between domains. However, it un-clear what factors determine whether - extent- useful. The long- standing assumption that features source get reused recently been called into question. Through series of experiments on several image bench-mark datasets, we explore...
Abstract Identifying active compounds for a target is time- and resource-intensive task in early drug discovery. Accurate bioactivity prediction using morphological profiles could streamline the process, enabling smaller, more focused compound screens. We investigate potential of deep learning on unrefined single-concentration activity readouts Cell Painting data, to predict across 140 diverse assays. observe an average ROC-AUC 0.744 ± 0.108 with 62% assays achieving ≥0.7, 30% ≥0.8, 7% ≥0.9....
The deep learning field is converging towards the use of general foundation models that can be easily adapted for diverse tasks. While this paradigm shift has become common practice within natural language processing, progress been slower in computer vision. In paper we attempt to address issue by investigating transferability various state-of-the-art medical image classification Specifically, evaluate performance five models, namely Sam, Seem, Dinov2, BLIP, and OpenCLIP across four...
Convolutional Neural Networks (CNNs) have reigned for a decade as the de facto approach to automated medical image diagnosis, pushing state-of-the-art in classification, detection and segmentation tasks. Over last years, vision transformers (ViTs) appeared competitive alternative CNNs, yielding impressive levels of performance natural domain, while possessing several interesting properties that could prove beneficial imaging In this work, we explore benefits drawbacks transformer-based...
High Content Imaging (HCI) plays a vital role in modern drug discovery and development pipelines, facilitating various stages from hit identification to candidate characterization. Applying machine learning models these datasets can prove challenging as they typically consist of multiple batches, affected by experimental variation, especially if different imaging equipment have been used. Moreover, new data arrive, it is preferable that are analyzed an online fashion. To overcome this, we...
Transfer learning is a standard technique to transfer knowledge from one domain another. For applications in medical imaging, ImageNet has become the de-facto approach, despite differences tasks and image characteristics between domains. However, it unclear what factors determine whether - extent useful. The long-standing assumption that features source get reused recently been called into question. Through series of experiments on several benchmark datasets, we explore relationship...
High content imaging assays can capture rich phenotypic response data for large sets of compound treatments, aiding in the characterization and discovery novel drugs. However, extracting representative features from high images that subtle nuances phenotypes remains challenging. The lack high-quality labels makes it difficult to achieve satisfactory results with supervised deep learning. Self-Supervised learning methods have shown great success on natural images, offer an attractive...
Abstract Efficiently identifying bioactive compounds towards a target of interest remains time- and resource-intensive task in early drug discovery. The ability to accurately predict bioactivity using morphological profiles has the potential rationalize process, enabling smaller screens focused compound sets. Towards this goal, we explored application deep learning with Cell Painting, high-content image-based assay, for prediction screening. Combining Painting data unrefined...
Evidence suggests that networks trained on large datasets generalize well not solely because of the numerous training examples, but also class diversity which encourages learning enriched features. This raises question whether this remains true when data is scarce - there an advantage to with additional labels in low-data regimes? In work, we consider a task requires difficult-to-obtain expert annotations: tumor segmentation mammography images. We show that, settings, performance can be...
The deep learning field is converging towards the use of general foundation models that can be easily adapted for diverse tasks. While this paradigm shift has become common practice within natural language processing, progress been slower in computer vision. In paper we attempt to address issue by investigating transferability various state-of-the-art medical image classification Specifically, evaluate performance five models, namely SAM, SEEM, DINOv2, BLIP, and OpenCLIP across four...
High Content Imaging (HCI) plays a vital role in modern drug discovery and development pipelines, facilitating various stages from hit identification to candidate characterization. Applying machine learning models these datasets can prove challenging as they typically consist of multiple batches, affected by experimental variation, especially if different imaging equipment have been used. Moreover, new data arrive, it is preferable that are analyzed an online fashion. To overcome this, we...