Probably this is a most common question a doctor would like to ask his/her colleague for several reasons. Probably they like to surprise, confuse them with a complicated case. May be they just like to start a conversation on the way to coffee or lunch. May be they have not figured out the case yet but would like some help. But my guess is mostly they have some a priory clues to which the friend is not privy. Whatever the reason may be one cannot deny that it is a provocative question for the friend or colleague had to respond.
I wonder how many of us have reflected on this question deeply. I mean philosophically or metaphysically, the question has deep meaning and consequences. What if the guy thinks he has cancer (or worse if he has one)? For such an important question why do we hardly spend any time on thinking about it? I meant epistemology of diagnosis. For sure many of us seniors are not or peripherally taught about ontology or epistemology in med school. Anyway med school professors hardly cared about teaching much less on philosophy which probably would be a constraint if one wants to make it big.
I guess much of what we learn in science is originated in west and hence has its roots in mono direction. What if logic is multidimensional and perception of our world differs in ways we think? If the way one thinks is molded by our cultures then definitely our perception of things would vary. If perception varies by our background then our inference also differ. So do we have different types of logic here? Am I talking about west vs Pandora as depicted in the epic Avatar?
Definitely there are different perceptions and hence any scientific pursuit should be open to all. If euclidean logic is sacrosanct and Socrates has the last say, why discovery of relativity was delayed? May be I am not an expert and may be I am wrong. But coming back to our question on diagnosis which was kept pending, I talk a bit on inference. According to Indian logic there are three types of inference: a). those that are concomitant in assertion and negation, b). those that are concomitant in assertion only and c). those that are concomitant in negation only. If a biopsy confirms or rules out a disease then the feature is type a. If brisk response to deep tendon reflexes suggest upper motor neuron lesion it is type b and if loss of deep tendon reflexes suggests lower motor neuron lesion it is type c. The patho-physiological processes here are important in diagnosis. For example presence or absence of hyper segmented nuclei is important in a biopsy for type a. Similarly release of inhibitory influence of upper motor neurons is in type b or loss of motor function in type c. So why nobody mentioned the logic in any medical text?
All of us are aware of Darwin’s theory of evolution. He classified all living things on earth systematically in so called ontological fashion. His theory on heredity is a logical derivation of parent and child classes in the classification. Description logic is based on first order predicate logic or euclidean logic. There are several reasoners like FACT++ or Pellet which are based on this and are designed to interpret and give automated inference. However, bio medical terminologies are too vast and complex and are not amenable for automated reasoning by the reasoners. So machine generated diagnosis is not possible because euclidean logic failed?
In a way, euclidean logic seems to have failed not because of its weakness but because of fragility of imagination on its premise. Intuitively what one perceives may be limited by what one can put in a equation leading to a fracture in our logic. If a child of a parent class inheres characteristics as defined in ontological principles, why not a tendon reflex (child) inheres some characteristics from central nervous system (parent) or brain? If yes, then features that are concomitant in assertion and negation are preferred higher than those that are concomitant only in assertion or negation. Extending same logic to disease (child), if its system (parent) or organ is shared by its feature, then it has higher preference to those that or not.
May be automated reasoning in diagnosis has limitations in terms of its accuracy. May be word vectors (as described by Google guys) may have real world implications to diagnosis. If so what are they? We read and write medical text in so many ways. What if we have a large corpus of medical text (say about 50 million words), and we have an equation that describe their contextual incidence in 360 degrees then each of our disease or its feature can be defined by a number. Implications of these numbers are astounding. It was found that solution for words (represented as numbers) in following equation is true: king – male + female = queen. If so then it might be true for disease and feature also. Then we have a way of hierarchical quantification which can lead us to an automated diagnosis.
Finally have we got a real working solution for automated reasoning? May be, may be not. It depends on what we are looking for. What I mean is whether we want a computer to diagnose cases for us or help us not miss a case. If we are looking for former, no and if you are looking for latter yes. So in summary we have a way of quantification of feature given disease in a hierarchically preferential fashion that we can use in a standard Bayesian probabilistic fashion to derive a diagnosis.