A system and a method of interpretable morphology-based hematologic diagnosis and risk scoring

A system (100) for morphology-informed hematologic diagnosis is disclosed. The system (100) comprises an imaging interface
(102) configured to receive a digital microscopic image of a biological specimen, a memory (106) storing a trained neural
network, predefined morphological descriptors, and clinically derived relevance weights, and a processor (104) operatively
coupled thereto. The processor (104) pre-processes the image using an image pre-processing engine (108), extracts morphology
features via a morphological feature extraction backbone (110), a multi-scale feature encoding module (112), and an attentionbased context integration module (114), and generates a latent representation through a unified feature embedding module (116).
The representation is projected into a morphology concept vector using a WHO-aligned concept bottleneck layer (118). A
differentiable clinical risk mapping layer (122) computes a scalar risk score, and a diagnostic classification head (120) generates
a diagnosis. A concept sensitivity scoring engine (126) and a counterfactual optimization module (128) provide explainability.
Fig. 1 is the reference figure.