RHINOS: A Consultation System for Diagnoses of Headache and Facial Pain

1. Input

RHINOS first requires that Information on the following 15 Items be screened.

Age Sex Pain location Nature of the pain Severity of the pain

History since onset

Frequency of the paroxysm

Duration of the paroxysm

Prodromal syndrome

Concurrent neurological sign during the paroxysm

Interval of the paroxysm

State of sleep

Ratio of present pain severity to the past severest period Tender spot

After acquiring information on these 15 signs, RHINOS applies rules for the case. If more information is required by the system, additional questions will appear on the screen.

2. Knowledge base

The knowledge base of RHINOS is mainly comprised of five kinds of knowledge, described earlier in this paper. The number of rules and images are as follows.

D-rule, SS-rule, WS-rule: 126

A-rule: 57

Disease-Image: 36

3. Inference engine

Procedure of inference sequence was described earlier in this paper (see II).

4. Output

Output is not a final diagnosis but a list of the possibilities from the view of anamneses and physical examinations. An example is given in fig. 2 .

V EVALUATION

Fifty cases of patients whose chief complaints was headache and/or facial pain were surveyed at Kansai Medical College, Department of Neurosurgery. A comparison was then made between the RHINOS diagnosis and that of a human expert (Professor of the department), whose diagnosis was based on the medical records of anamneses and physical examinations only.

In 40 out of 50 cases, both the RHINOS' and the doctor's diagnoses coincided. In the 1 0 other cases the following problems arose:

Type 1. RHINOS did not rule out a disease the expert ruled out. (5 cases)

Type 2 . RHINOS stated "This case may be disease. d1, but manifestation mj can't be explained." The expert rejected this possibility. (2 cases)

Type 3. RHINOS stated only one disease with "manifestation m j can't be explained" The expert concluded that the case was a complications of two diseases. (1 case)

Type 4. RHINOS could not provide conclusive information because the disease was very rare and no existing rule applied to that disease. (1 case)

Type 5. The medical record was apparently erroneous. (1 case)

Actually, there is no serious problem in Type 1. Even in Types 2 and 3, RHINOS provided a suggestion for the right answer. Discarding Type 5, the authors reacned the following conclusions:

tttttttttttttttt RHINOS Dia~nostic Conclusions +tttttttttttttt++++++++++·

tl ARTERIOSCLEROTIC HEADACHE is stroMlY suspected , tl

l!ecaus@

History since onset Nature of Pain

Sclerosis of retinal artery Pain a~~revated

Furthermore, Tenderness

Jolt headache

: WavY continuous

: Continuous Pain

Diurnal variation

Yes

EsPeciallY in the aornins,

Al ( ri~htl <definite Pain)

Al <left) (definite Pain)

: Answer Yes Just after Joltin~

these manifestations stren~then the certainty of this di·a~nostic belief,

• tl There is sooe Possibility of INTRACRANIAL HASS LESION. tl

Because

Jolt headache

: Answer yes Just after Joltin~.

But this dia~nosis cannot exPlain the following oanifestations: Sclerosis of retinal arterY : Yes.

• U There is some PossibilitY of SYHPTOHS OF OTHER DISEASES, tt

Because

Ratio or Present severity of Pain compared .to Past severest Period

HistorY since onset Jolt headache

Further·ore' Tenderness

: Wavy contiuous

: Here than 5/10

: Answer Yes Just after Joltins.

: Al <riShtl <definite Pain)

Al (left) (definite Pain) these ·ani festations strensthen the certaintw of this diasnostic belief,

But this dia~nosis cannot exPlain the following unifestations: Sclerosis of retinal arterY : Yes,

Following diseases must be differentiated: AUTONOMIC DISTURBANCE

Coincided with the expert: 82% Answered with little discordance: 16% Discordance: 2%

The result show that RHINOS is capable of providing almost the same as a human expert. However, the following problems still remain.

1. The disease-image of RHINOS is a set of manifestations of a disease that can arise, but the disease image of a human expert is far more complex. This made it impossible for RHINOS to rule out a disease that the expert ruled out.
2. 2 . Because manifestations of headache and/or facial pain are rather subjective, a certain amount of ambiguity must be accouted for in the information obtained. The human expert put emphasis on the key manifestations, which he then recon­ firmed in many ways in order to get the information as correct as possible. RHINOS does not do this.

VI FINALING REMARKS

RHINOS is now used mainly by the resident staff in the office of the Department of Neurosurgery. It is also used as CAl for students. The portab!Uty of the system, as well as the quality of consultation, enhanced the actual use of this sys­ tem.

As noted before, In order for this system to make more pre­ cise diagnoses for patients suffering from more than one disease, the introduction of classifications of disease-image (necessary image, sufficient image, exclusionary image, etc.) will prove effective in diagnosis.

The authors are already planning some other improvements on the system. It is planned for RHINOS to focus on and reconfirm key manifestations that are fundamental to diag­ nosis.

The authors are already planning some other improvements on this sytem. It is planned for RHINOS to focus on and recon­ firm key manifestations that are fundamental to diagnosis. Plans also exists for RHINOS to suggest, 1) what laboratory examinations be examined in order that a more precise diag­ nosis may be achieved; and, 2) methods of treatment beyond diagnosis.

In this project, the authors confirmed their hypothesis to the domain of headache diagnosis, even though it was a subjective-fact oriented domain. Being so close to the funda­ mental procedure of doctors' diagnoses, this hypothesis can be applied to other diagnostic domain as well.