Computational Hematology |

Digital staining of white blood cells with confidence estimation

Sun, 01 Jan 2023 00:00:00 +0000

Anomaly-aware multiple instance learning for rare anemia disorder classification

Sat, 01 Jan 2022 00:00:00 +0000

Unsupervised Cross-Domain Feature Extraction for Single Blood Cell Image Classification

Sat, 01 Jan 2022 00:00:00 +0000

What can we learn about a generated image corrupting its latent representation?

Sat, 01 Jan 2022 00:00:00 +0000

Fourier Transform of Percoll Gradients Boosts CNN Classification of Hereditary Hemolytic Anemias

Fri, 01 Jan 2021 00:00:00 +0000

Sickle Cell Disease Severity Prediction from Percoll Gradient Images using Graph Convolutional Networks

Fri, 01 Jan 2021 00:00:00 +0000

Multi-task multi-domain learning for digital staining and classification of leukocytes

Mon, 21 Dec 2020 00:00:00 +0000

AI for Hematology: Trustworthy digital staining and classification of white blood cells

Wed, 01 Jan 2020 00:00:00 +0000

Chemical staining of blood smears is a routine but expensive, slow and sensitive step of blood analysis, and the lack of unified protocols makes colours and visible structures vary between laboratories. Deep generative models can replace the chemical process with a digital one — but in a medical setting a convincing synthetic image is not enough: it has to keep the cell’s diagnostic structures, and the system has to know when its output cannot be trusted.

What we did

Digital staining that preserves the cell. A multi-task, multi-domain generative network translates unstained into stained white-blood-cell images while learning nucleus and cytoplasm segmentation and direct reconstruction as auxiliary tasks, so that inter-cellular structures survive the translation. Injecting the target-domain label into the generator gives a domain-agnostic latent space that also supports automated leukocyte classification. Validated on leukocytes from 24 patients, with state-of-the-art staining and classification performance ( ).
Reading reliability from the latent space. Corrupting a GAN’s bottleneck representation with noise and comparing the resulting outputs reveals how robust that representation is — predicting uncertain regions of a synthetic image and flagging samples unfit for downstream tasks ( ).
Confidence for every stained cell. Disentangling the latent space into separate representations of the cell and of the stain yields separate confidence estimates for structure and staining quality; the estimated confidence correlates with perceptual image quality (LPIPS) and identifies unreliable synthetic cells ( ).

Our role

Methodological lead on generative modelling and uncertainty estimation, and co-supervision — with Slobodan Ilic (Siemens AG) and Nassir Navab (TU Munich) — of Agnieszka Tomczak’s industrial PhD, completed at TU Munich in 2023.

Perspective

The project turned digital staining from an image-translation exercise into a question of trust, and laid the groundwork for our later work on uncertainty estimation and reliable medical image generation.

Attention Based Multiple Instance Learning for Classification of Blood Cell Disorders

Wed, 01 Jan 2020 00:00:00 +0000

Explainable AI for rare blood disorders

Tue, 01 Oct 2019 00:00:00 +0000

Rare anaemias such as hereditary spherocytosis or sickle cell disease are diagnosed from blood samples in which only a fraction of the red blood cells is visibly altered. Labelling every cell is laborious and subjective, specialised genetic tests are expensive and unavailable in most labs, and for rare diseases there are simply very few patients to learn from. Clinicians also need to see why a model reaches a diagnosis before they can trust it. Together with Carsten Marr’s group at Helmholtz Munich and haematology partners (Anna Bogdanova, University of Zurich), we developed explainable methods that diagnose at the level of the patient while pointing to the individual cells that matter.

What we did

Attention-based multiple instance learning. Each blood sample is treated as a bag of single red blood cells detected with an R-CNN; an attention-based MIL model classifies the patient into one of four blood cell disorders, and the attention weights show how much each cell contributed — improving both accuracy and interpretability for the haematologist ( ).
Percoll density gradients as an affordable biomarker. Centrifuging a patient’s blood through a Percoll gradient separates red cells by density. Combining CNN features with Fourier spectral features of the gradient images classified hereditary haemolytic anaemias with an average F1-score of 88% ( ), and graph convolutional networks on the gradient profiles predicted sickle cell disease severity ( ).
Robustness across labs. A cross-domain adapted autoencoder extracts white blood cell features in an unsupervised way, so that classifiers trained on one dataset transfer to smears from other sites, stains and microscopes ( ).
Anomaly-aware MIL for very rare disorders. By using healthy cells from negative bags as a reference, an interpretable pooling method raises the contribution of anomalous cells and outperforms attention-based MIL on rare anaemia classification with few training samples ( , with Salome Kazeminia).

Our role

Co-supervision of Ario Sadafi’s PhD (TU Munich and Helmholtz Munich, with Carsten Marr and Nassir Navab), started at TU Munich CAMP and continued at Helmholtz AI, and contributions to the work of Salome Kazeminia and Raheleh Salehi in Carsten Marr’s group.

Perspective

The project showed that patient-level labels are enough to train cell-level explanations, a principle we now apply to weakly supervised learning in computational pathology and other rare-disease settings.