Treatment Regimen Segmentation from Handwritten Medical Prescriptions Using Advanced Neural Network

Abstract

Handwritten medical prescriptions are a critical yet under-digitized component of clinical workflows, often serving as a source of ambiguity due to illegible handwriting, overlapping text blocks, and structural inconsistencies. The automatic segmentation of such prescriptions into meaningful textual blocks is vital for downstream tasks like drug recognition and dosage extraction. Traditional methods grounded on connected components or projection profiles often falter under the irregularities of freeform handwriting. To address these limitations, the paper proposes an advanced deep learning architecture—PrescNet—that primarily segment the treatment regimen (medicine and its associated components) as text-blocks using classical U-Net design with spatial–channel attention gates and a lightweight 32 channel projection layer to better capture salient features in prescription images. The model is trained on a custom dataset with pixel-level annotations and evaluated using 10-fold cross-validation with varying data splits. Experimental results demonstrated that the proposed architecture significantly transcend the baseline variants and a few state-of-the-art deep learning models of text-line segmentation achieving an Intersection over Union (IoU) of 87.2%, Dice score of 92.9%, and a minimal Dice loss of 0.071. The results validate its effectiveness in handling complex handwritten layouts, establishing its suitability for real-world clinical applications.

Received: 12 September 2025 | Revised: 30 December 2025 | Accepted: 14 January 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

Data available on request from the corresponding author upon reasonable request.

Author Contribution Statement

Rekha G. R.: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Siddesha S.: Conceptualization, Validation, Formal analysis, Investigation, Writing – review & editing, Supervision, Project administration. V. N. Manjunath Aradhya: Validation, Formal analysis, Investigation, Writing – review & editing, Supervision.