AI and Rare Diseases ── Rare blood disorders ── 2026-10-16

Blood smear images are the input. The challenges are image noise, varied RBC morphology and overlapping cells. The Medav-Gaussian filter denoises, Tsallis entropy-based thresholding segments cells, and images are augmented. Geometric and first-order statistical features are extracted and PShCN-Net classifies. Tested on 90% of training data, accuracy is 91.945%. There is no external validation; multi-center validation is future work. — Image abstract — the whole article on one page (click to enlarge)

1 Rare blood disorders papers. Lead: Sickle cell disease classification using Medav-Gaussian filter and Pyramid Shepard CNN

Journals covered and how papers are chosen: see the index

Rare blood disorders: treatment, trials and AI

Sickle cell disease classification using Medav-Gaussian filter and Pyramid Shepard CNN
Scientific Reports · 2026-10-13 · Evidence: Diagnostic model development study (no external validation)
Abstract
Sickle cell disease (SCD) poses serious health risks and requires accurate classification for effective treatment. In SCD, red blood cells (RBCs) mutate into a sickle shape due to a haemoglobin gene mutation, impairing oxygen delivery and causing blockages in blood vessels. However, accurate SCD classification from blood-smear images remains challenging due to image noise, variations in RBC morphology and overlapping cells. To address these challenges, this study introduces a deep learning (DL)- based approach named Pyramid Shepard convolutional neural network (PShCN-Net) for classifying SCD using blood smear images. The process begins with image denoising using the Medav-Gaussian (MG) filter, followed by cell segmentation with Tsallis entropy-based thresholding. Image augmentation is applied to enhance model robustness, and geometric and first-order statistical features are extracted. Classification is then performed using the proposed PShCN-Net, which combines the PyramidNet and Shephard convolutional neural network architectures for improved accuracy. PShCN-Net attains 91.945% accuracy, 93.235% true positive rate, 92.345% true negative rate, 91.437% precision, 92.327% F1-score, 7.655% false positive rate and 6.765% false negative rate when tested on 90% of the training data. Future work will focus on validating the PShCN-Net using larger and diverse multi-center blood-smear datasets to further assess its generalizability and clinical applicability.
Journal IF-equivalent: 5.2 (OpenAlex 2-year mean citedness, value as of 2026-10-06, retrieved 2026-10-07; not the official Clarivate IF)
Reference: Ponnada S, Majji R, Rout AK, Selvaraju S. Sickle cell disease classification using Medav-Gaussian filter and Pyramid Shepard CNN. Sci Rep. 2026 Oct 13 [Epub ahead of print]. doi:10.1038/s41598-026-75109-8.
Checked: Abstract only

← Back to this issue