Indmedica - India's Premier Medical Portal - Cyber Lectures


Basic Medical Science Education Must Include Medical Informatics

Suptendra Nath Sarbadhikari

Page 3


Rule Generation:

The trained MLP is used for rule generation in If-Then form. These rules describe the extent to which a pattern belongs or not to one of the classes in terms of antecedent and consequent clauses. For this, one has to backtrack along maximal weighted paths using the trained net and utilize its input and output activations. After training, the connection weights of an MLP encode among themselves (in a distributed fashion) all the information learned about the input-output mapping. Therefore, any link weight with a large magnitude reflects a strong correlation between the connecting neurons. This property is utilized in evaluating the importance of an input node (feature) on an output layer mode (decision). For this one needs to compute the path weights from each output node to each input node through the various hidden nodes. Each output-input path that is maximum (through any hidden node) denotes the importance of that input feature for arriving at the corresponding output decision. Note that one or more hidden layers may be considered (if necessary) for evaluating the path weights. An increase in the number of hidden layers simply leads to an increase in the possible variety of paths being generated through the various hidden nodes in the different layers. For the details the interested reader may kindly refer to Sarbadhikari and Pal4.

Computers in medical training

Computer aided or assisted instruction (CAI) is a tutorial method using a computer as a base for managing the student’s progress (at the desired pace and time). Computer aided or assisted learning (CAL) is a computer based tutorial method that uses the computer to pose questions, provide remedial information and chart a student through a course. Especially now with the emphasis, especially in medical related fields, on problem based learning (PBL), these have gained greater significance. In the case of PBL, the approach is top down, rather than bottom up, i.e., instead of starting from Physiology or Biochemistry, one discusses a patient’s presenting features and goes backwards to find the physiological and biochemical basis of the disease presented. Here also electronic tutorials can be of great help to the students. Especially in the cases where some invasive procedures have to be used, ‘simulators’ are far better suited since they do not cause pain or harm to any real person. Cadaver dissection does not give an idea of the functional correctness of the organ structures, while animal physiological studies give a wrong impression of the human organs’ anatomy. While studying on patients is commonly done (because of the population explosion in our country there is certainly no dearth of patents!) it may not be ethically sound in all the cases. The other common practice of trying things out on each other (classmate or roommate) is also not without its share of pain, discomfort, and occasionally shyness.

Evidence based medicine (EBM)

For selecting the appropriate alternative regime for a particular patient, there is no definitive method. Data sheets on various medicines purveyed by the pharmaceutical manufacturers are not of much help since they focus on only one particular (generic or brand) drug, while doctors require comparative data on all available treatment modalities. To overcome this problem, Evidence Based Medicine (EBM) compiles data based on systematic review of RCTs (randomized control trials), through ‘meta-analysis’, involving all available therapies. To cite a concrete example, the comparative effectiveness of various interventions in relieving or reducing pain of osteoarthrosis (OA) can be summarized as: Beneficial and likely to be beneficial. In the first group we get systemic simple analgesics like paracetamol (acetaminophen) for short-term pain relief and improvement in function, systemic NSAIDs (non steroidal anti-inflammatory drugs), which too give short-term pain relief and functional improvement and topical analgesic agents (for short term pain relief). In the second group we find education, dietary advice, empowerment, and support (improved knowledge of the disease and pain relief). Physical support also purveys pain relief and functional benefits. The key points to note are: no good evidence that NSAIDs are superior to paracetamol or to suggest that any one of the various NSAIDs available in the market has greater efficacy in pain reduction in OA. One systematic review of RCT has found that topical agents relieve pain in OA and offer a less toxic alternative to systemic drug therapy. However, there is no evidence to the effect that prescribed local analgesics are superior to cheaper, non-prescription OTC (over the counter) alternatives or local hot/cold packs. On recovery from active inflammation, a program of stepwise simple (isotonic) exercise (like quadriceps exercise or stretching and relaxing the knee joints in knee OA) is often essential.

Case Based Reasoning (CBR)

A medical practitioner encountering a new problem is usually reminded of the similar cases seen in the past. New problems are solved by analogy with the old ones and explanations and reasoning are often derived from the prior experiences. This sort of a problem solving, by computer systems is called case based reasoning or CBR. It must have a large case library instead of a set of rules. The indexing system must also be of very high quality. Albeit this is a computer learning paradigm, in the case of medical expert systems, this problem solving approach is often very useful. Here the learning may combine both inductive (several positive and negative examples) or explanation based learning (EBL), where a single example is enough to generalize a solution. For clinical decision making, iterative dichotomizers (ID) or decision tree form of inductive learning and EBL methods may be used by computers.

Such a system must find out: (A) how cases are organized in memory, (B) how are relevant cases retrieved from memory, © how can previous cases be adapted to new problems and (D) how cases are originally acquired. For organizing cases in memory, a rich indexing system must be used. For retrieving and matching the best case from a huge database of cases, preference heuristics may have to be used. They may be:

  1. Goal-directed preference: preferring cases that involve the same goal as current case.
  2. Salient-feature preference: preferring cases that match the most important or maximum number of features.
  3. Specificity preference: preferring cases that match features exactly rather than generally.
  4. Frequence preference: preferring frequently matched cases.
  5. Recency preference: preferring recently matched cases.
  6. Ease-of-adaptation preference: preferring cases with features that are easily adapted to new situations.

As the best case will also not match the current situation exactly, it has to be adapted. This can be done by mapping new objects into old ones. For this, the case library may have to be augmented with a plan-modification library and another plan-repair module for future reference. Finally, for acquiring original cases, a thorough CBPR is absolutely necessary.

1 2 3 4 5 Previous page Next page Back