AI can read medical certificates in a variety of formats in just 3 seconds

We have been providing Cinnamon AI's original high-precision AI-OCR engine, "Flax Scanner," to Meiji Yasuda Life Insurance Company (hereinafter, "Meiji Yasuda") for the digitization of "medical treatment statements" and "prescription statements." We are now pleased to announce that starting in November, we have expanded the scope of our services to include the reading of "health certificates."

Meiji Yasuda handles various types of documents issued by medical institutions in its daily operations. Accurately reading specialized text from a huge number of documents is a labor-intensive and highly personal task, which is costly and an issue. The expansion of the use of "Flax Scanner," which can read non-standard forms, will enable the company to focus on core operations, and will contribute to further improving its services and promoting digital transformation, such as strengthening emergency response capabilities through the introduction of AI.

■ AI quickly and accurately digitizes 91 items on a health certificate in about 3 seconds per sheet

The AI-OCR "Flax Scanner" that Cinnamon AI will provide to Meiji Yasuda has been tuned to be able to quickly and accurately read health certificates in various formats issued by various medical institutions. In particular, the AI automatically digitizes 91 frequently appearing items on health certificates with an average item-by-item accuracy of 89.061 TP3T, and 26 items designated as priority items with an average accuracy of 90.691 TP3T. In addition to scanned data from health certificates, it can also read image data taken with cameras such as smartphones, which are prone to data quality variations due to camera shake, shadows, and tilt. The reading speed per data sheet is approximately 3 seconds, significantly reducing the manual data entry work that has been required up until now.

■ Cinnamon AI's original high-precision AI-OCR "Flax Scanner"

We have developed and are providing our original high-precision AI-OCR engine, "Flax Scanner," which allows AI to understand the meaning and read data from documents in various formats in any business scene, just like a human. Conventional AI-OCR mainly uses "coordinate definition type" that defines the coordinates in advance for formatted documents, but it is difficult to handle documents such as invoices, which have countless formats for each company. In contrast, our "Flax Scanner" has excellent technology for "feature learning type" that does not require prior coordinate definition, and is capable of reading a variety of forms and documents with high accuracy.

Furthermore, it can also extract information from industry-specific technical terms, drawings, and difficult photographic images, and has received high praise in this technical field, known collectively as IDP (Intelligent Document Processing, or automatic knowledge extraction technology), with papers being accepted at international conferences.

In addition, a new, highly versatile AI-OCR platform called "Flax Scanner HUB" has been launched, which allows easy use of Cinnamon AI's high-precision AI-OCR "Flax Scanner" for business purposes. "Flax Scanner HUB" is equipped with three different AI engines: coordinate definition type, feature learning type, and generative AI extraction type, and is capable of extracting data with high accuracy even from non-standard forms that are difficult to read. The feature learning type and generative AI extraction type do not require prior coordinate definition, allowing a variety of forms to be used in one stop.

Flax Scanner HUB:https://cinnamon.ai/flax-scanner-hub/　

<Reference materials>

 Cinnamon AI High-precision AI-OCR "Flax Scanner"
Meiji Yasuda's "health certificate" - 91 items to be read

■ Reading accuracy (test data values by Cinnamon AI)

Accuracy of key items Character unit accuracy: Average 94.391 TP3T Accuracy by item: Average 90.69%
Accuracy of all items Character unit accuracy: Average 92.82% Accuracy by item: Average 89.06%

*26 important reading points

No.	Reading item name	No.	Reading item name
1*	Last name	26	Fundus examination (Scheie classification: h left)
2*	given name	27	Fundus examination (Scheie classification: s left)
3*	Kana (surname)	28	Fundus examination (Scott classification, left)
4*	Kana (Name)	29*	Maximum blood pressure 1st time
5*	Name of medical institution	30*	Diastolic blood pressure 1st time
6*	Insurer information	31*	Maximum blood pressure 2nd time
7*	Date of visit	32*	Diastolic blood pressure 2nd time
8*	height	33	Chest x-ray
9*	body weight	34	Electrocardiography
10*	BMI	35	Red blood cells
11	Waist circumference	36	Hemoglobin (Hb)
12	Right hearing (1000Hz)	37	GOT (AST)
13	Right hearing (4000Hz)	38*	GPT (ALT)
14	Hearing left (1000Hz)	39*	γ-GTP
15	Hearing left (4000Hz)	40	HDL
16	Hearing Speech Range	41	LDL
17	Vision (right)	42*	Neutral fats
18	Vision (left)	43*	Fasting blood glucose
19	Vision (right) correction	44*	Urine sugar semi-quantitative
20	Vision (left) correction	45*	Urine sugar (quantitative)
21	Fundus examination (Keith Wagner classification, right)	46*	Urine sugar (qualitative)
22	Fundus examination (Scheie classification: h right)	47*	Urinary protein semi-quantitative
23	Fundus examination (Scheie classification: s right)	48*	Urinary protein (quantitative)
24	Fundus examination (Scott classification right)	49*	Urinary protein (qualitative)
25	Fundus examination (Keith Wagner classification, left)	50	Urine occult blood

No.	Reading item name	No.	Reading item name
51	Urinary sediment (red blood cells)	76	Pulmonary function test (forced vital capacity)
52	Urinary sediment (white blood cells)	77	Pulmonary function test (forced expiratory volume in one second)
53	Urinary sediment (squamous epithelium)	78	Pulmonary function test (forced expiratory volume in one second)
54	White blood cells	79	Pulmonary function test (% forced expiratory volume in 1 second)
55	Hematocrit	80	Pulmonary function test (% vital capacity)
56	platelet	81	Stomach X-ray
57	Blood picture (neutrophils)	82	Abdominal ultrasound
58	Blood picture (eosinophils)	83	Red blood cell constants (MCV)
59	Blood picture (basophils)	84	Red blood cell constants (MCH)
60	Blood picture (monocytes)	85	Red blood cell constants (MCHC)
61	Blood picture (lymphocytes)	86	HBsAg
62	Total Protein	87	HCV antibodies
63	albumin	88	Tumor marker (PSA)
64	Total bilirubin	89	Breast Cancer Screening
65	amylase	90	Cervical cancer screening
66	LDH	91*	Random blood glucose
67	TC (total cholesterol)
68	Non-HDL cholesterol
69*	HbA1C (NGSP)
70	Serum creatinine
71	eGFR
72	uric acid
73	Stool occult blood (day 1)
74	Stool occult blood (2nd day)
75	Pulmonary function test (lung capacity)

▼Contact information for companies regarding this matter
https://contents.cinnamon.ai/contact/inquiry