Verified, then logged.
Sun 22 Nov 2026 · 00:33 KST / 15:33 UTC한국어 | ENSearch /

Breaking01 / AI

Phone prescription reader reports 67% medication recall

A study using the same prescription photographs reported fewer missed medications but longer processing times.

AI-generated editorial illustration for this article

Researchers reported 67% medication recall for a phone-based prescription reader when they tested an updated system on the same photographs as its predecessor.

The paper, published on arXiv on 17 Nov, describes an evaluation using 49 photographs of 15 synthetic prescriptions. One person wrote the handwriting, and testing used one device. Recall measures how completely a system identifies prescribed medications.

According to the paper, the original system combines PaddleOCR, a text-recognition tool, with a 270M-parameter language model. The updated system crops rotated lines of text and uses a 2B-parameter vision-language model to read them again. It also checks numbers against each line.

In the researchers' published figures, recall rose from 8% to 67%. The number of handwritten medications detected increased from 0 to 20 out of 45.

The researchers reported that processing time per photograph increased from about 12 seconds to 61 seconds. The updated system uses about 1.9 GB of memory.

The researchers attributed much of the poor handwriting performance to a defect in their own image-cropping step. They identified the procedure that accepts or rejects text based on text-recognition confidence as the main remaining limitation.

The researchers have not published evaluation results involving other writers or devices.