Characterization and Classification of Brain Tissue and Stroke Lesions in Non-Contrast Computed Tomography Images of Stroke Patients Using Statistical Texture Descriptors and Artificial Neural Network

Q: What are the key findings of this study?

Histogram texture descriptors with an artificial neural network achieved 0.947 sensitivity, 0.962 specificity, and 0.980 AUROCC in classifying brain tissue and stroke lesions on CT (p<0.05).

Q: What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: Stroke (ischaemic and haemorrhagic) (n=164). Intervention: Artificial neural network classification using histogram statistical texture descriptors vs. Co-occurrence matrix, run-length matrix, and absolute gradient texture descriptors (validated against radiologists' visual categorization). Primary outcome: Classification sensitivity, specificity, and area under the ROC curve (AUROCC) for discriminating normal brain tissue, ischaemic stroke, and haemorrhagic stroke lesions (p=< 0.05).

Q: What question did this study set out to answer?

To characterize and automatically classify normal brain tissue, ischaemic stroke, and haemorrhagic stroke lesions in non-contrast CT images using statistical texture descriptors and an artificial neural network.

ph. D Ohagwu; Ph. D Agwu; B Sc Onyekelu; FMCR Mohammad; FMCR Mohammad; M Sc Abba

doi:10.82547/jrrs.vol32no1.69

December 31, 2025Open Access

Characterization and Classification of Brain Tissue and Stroke Lesions in Non-Contrast Computed Tomography Images of Stroke Patients Using Statistical Texture Descriptors and Artificial Neural Network

View Full Paper

Ask AI

Bookmark

Share

Key result

Neural network CT texture analysis classifies stroke lesions with ~98% AUROCC.

P<0.05
n=164

Key Points

To characterize and automatically classify normal brain tissue, ischaemic stroke, and haemorrhagic stroke lesions in non-contrast CT images using statistical texture descriptors and an artificial neural network.
Analyzed non-contrast CT scans of 164 stroke patients, where two blinded radiologists categorized lesions as ischaemic or haemorrhagic to serve as the gold standard.
Extracted four regions of interest per slice (two lesion, two normal tissue) and computed statistical texture descriptors: co-occurrence matrix, run-length matrix, absolute gradient, and histogram parameters.
Trained an artificial neural network to classify regions of interest and evaluated diagnostic performance using receiver operating characteristic curve analysis.
Histogram-derived texture features yielded the highest artificial neural network diagnostic performance, achieving a sensitivity of 0.947, specificity of 0.962, false positive rate of 0.038, false negative rate of 0.053, and an overall accuracy of 94.68% (p < 0.05 compared to other texture classes).
Co-occurrence matrix features produced a sensitivity of 0.637, specificity of 0.753, false positive rate of 0.247, and false negative rate of 0.363.
Run-length matrix features achieved a sensitivity of 0.544, specificity of 0.607, false positive rate of 0.393, and false negative rate of 0.456, while absolute gradient features showed a sensitivity of 0.546, specificity of 0.586, false positive rate of 0.414, and false negative rate of 0.454.

Study Design

Type

Cross-Sectional (n=164)

Blinding

Blinded radiologists

Multicenter

Yes

PICO

P

Population

164 patients (mean age 60 years, range 32-85) clinically diagnosed with ischaemic or haemorrhagic stroke confirmed by two blinded radiologists on non-contrast head CT across two Nigerian centers.

E

Exposure / Comparator

Artificial neural network classification using histogram statistical texture descriptors vs Co-occurrence matrix, run-length matrix, and absolute gradient texture descriptors (validated against radiologists' visual categorization)

O

Primary Outcome

Classification sensitivity, specificity, and area under the ROC curve (AUROCC) for discriminating normal brain tissue, ischaemic stroke, and haemorrhagic stroke lesions, p=< 0.05

Main Result

Absolute Event Rate: 0.947% vs 0.637%

p-value: p=< 0.05

Limitations

Method for identifying mistaken classification cases by the computer-aided scheme was not considered
Clinical mimics of stroke were not included, so ability to distinguish stroke from mimics could not be assessed
Post-ictal intervals before CT imaging were not captured, preventing evaluation of time-dependent changes in CT lesion appearance

Implication

Radiologists identify stroke lesions in brain tissue, revealing the effectiveness of AI in classification.

Abstract

Patients and methods: Two experienced radiologists blinded to each other inspected CT images of 164 stroke patients to identify and categorize stroke lesions into ischaemic and haemorrhagic subtypes. Four regions of interest (ROIs) in each CT slice that demonstrated the lesion; two each representing the lesion and normal tissue were selected. Statistical texture descriptors namely, co-occurrence matrix, run-length matrix, absolute gradient and histogram were calculated for them. Raw data analysis was performed to identify the parameters that best discriminate between normal brain tissue and stroke lesions. Artificial neural network (ANN) was used to classify the ROIs into normal tissue, ischaemic and haemorrhagic lesions using the radiologists’ identification and categorization as the gold standard, and further analyzed using the receiver operating characteristic curve. Results: Three parameters in each texture class discriminated between normal tissue, ischaemic and haemorrhagic stroke lesions. The discriminating co-occurrence matrix parameters were sum average parameters namely S1-1 SumAverg, S1-0 SumAverg and S0-1 SumAverg. For the run-length matrix, short run emphasis in horizontal, 1350 and 450 directions were the discriminating features. The discriminating absolute gradient parameters were gradient non-zeros, gradient variance and gradient mean. For the histogram class, the mean, 90th and 99th percentiles were the discriminating parameters. The ANN achieved a sensitivity of 0.637, specificity 0.753, false positive rate (FPR) 0.247, and false negative rate (FNR) 0.363 with co-occurrence matrix. With run-length matrix the sensitivity was 0.544, specificity 0.607, FPR 0.393, and FNR 0.456 while with absolute gradient the sensitivity was 0.546, specificity 0.586, FPR 0.414, FNR 0.454. With histogram, the sensitivity was 0.947, specificity 0.962, FPR 0.038, and FNR 0.053. Conclusion: The histogram texture features showed the highest sensitivity and specificity in the classification of brain tissue and stroke lesions using the artificial neural network.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ohagwu et al. (2025) conducted a cross-sectional in Stroke (ischaemic and haemorrhagic) (n=164). Artificial neural network classification using histogram statistical texture descriptors vs. Co-occurrence matrix, run-length matrix, and absolute gradient texture descriptors (validated against radiologists' visual categorization) was evaluated on Classification sensitivity, specificity, and area under the ROC curve (AUROCC) for discriminating normal brain tissue, ischaemic stroke, and haemorrhagic stroke lesions (p=< 0.05). Histogram texture descriptors with an artificial neural network achieved 0.947 sensitivity, 0.962 specificity, and 0.980 AUROCC in classifying brain tissue and stroke lesions on CT (p<0.05).

synapsesocial.com/papers/6980fe57c1c9540dea810457 https://doi.org/10.82547/jrrs.vol32no1.69

View Full Paper

Ask AI

Bookmark

Share

Characterization and Classification of Brain Tissue and Stroke Lesions in Non-Contrast Computed Tomography Images of Stroke Patients Using Statistical Texture Descriptors and Artificial Neural Network

Key Points

Study Design

PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Characterization and Classification of Brain Tissue and Stroke Lesions in Non-Contrast Computed Tomography Images of Stroke Patients Using Statistical Texture Descriptors and Artificial Neural Network

Key Points

Study Design

PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading