DOC PREVIEW
MIT HST 950J - Medical Natural Language Processing

This preview shows page 1-2-17-18-19-35-36 out of 36 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 36 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Medical Natural Language Processing 6.872/HST950The Dream • Develop a comprehensive, precise language of expression for all clinical data • It’s the language that is precise.Thus, it must be able to state imprecision, uncertainty, etc. • Translate all actual clinical text into this language • Develop reasoning/inference methods to draw consequences within this language • Get clinicians (and others) to use thisThe Reality • Most clinical records of observations, interpretations and procedures are stated in free-form natural language • There are many sources of error and ambiguity • Language is infinitely varied • Computers are still poor at doing most text analysis tasks • But, with significant exceptions, especially for narrow tasks • Different approaches work best for different tasks -- no universal methodsStructure in a CHB ED Note Patient seen: 11:45 AM 21 year old male patient injured his right knee. The injury occurred when he was tackled while playing football 2 days ago. He complains of pain and swelling along the medial aspect of the right patella, medial collateral ligament of the right knee and medial collateral ligament of the right knee. He has been able to bear weight. His symptoms are exacerbated by bending his knee.. He has used a knee immobilizer. With some relief. CURRENT MEDICATIONS: None. ALLERGIES: Denies known allergies. IMMUNIZATIONS: Up to date. PE: Alert. In no acute distress. Well-developed. Well-nourished. RightKnee: Positive for tenderness and swelling involving the medial condyle of the distal right femur. There is no effusion or ecchymosis. Full range of motion. Slight limp. Normal bulk, tone, and strength. Sensation intact. The examination of the other knee is unremarkable. There is no evidence other trauma. Other PE: No other injuries. TREATMENT & COURSE: Knee immobilizer applied. DISPOSITION/PLAN: Discharged in good condition. ASSESSMENT: 1. Sprain of the medial collateral ligament 844.1. ATTENDING NOTE: Discussed with me agree with plan. 4Bulk of Valuable Data are in Narrative Text Mr. Blind is a 79-year-old white white male with a history of diabetes mellitus, inferior myocardial infarction, who underwent open repair of his increased diverticulum November 13th at Sephsandpot Center. The patient developed hematemesis November 15th and was intubated for respiratory distress. He was transferred to the Valtawnprinceel Community Memorial Hospital for endoscopy and esophagoscopy on the 16th of November which showed a 2 cm linear tear of the esophagus at 30 to 32 cm. The patient’s hematocrit was stable and he was given no further intervention. The patient attempted a gastrografin swallow on the 21st, but was unable to cooperate with probable aspiration. The patient also had been receiving generous intravenous hydration during the period for which he was NPO for his esophageal tear and intravenous Lasix for a question of pulmonary congestion. On the morning of the 22nd the patient developed tachypnea with a chest X-ray showing a question of congestive heart failure. A medical consult was obtained at the Valtawnprinceel Community Memorial Hospital. The patient was given intravenous Lasix. A arterial blood gases on 100 percent face mask showed an oxygen of 205, CO2 57 and PH 7.3. An electrocardiogram showed ST depressions in V2 through V4 which improved with sublingual and intravenous nitroglycerin. The patient was transferred to the Coronary Care Unit for management of his congestive heart failure , ischemia and probable aspiration pneumonia.Some Typical Tasks • Information retrieval -- usually, find an article relevant to x • Question answering -- answer specific questions from information represented in text • Learn and generalize -- find and categorize all protein-protein interactions reported in research literature • Case selection -- find patients based on their clinical characteristics; e.g., find asthmatics who don’t smoke • Extract diagnoses, symptoms, tests, results, medications, outcomes, etc., from clinical records • Extract relations among the above: e.g., x was done to rule out y • Find (and suppress) identifying information to make data safe for public releaseMethods • grep • Search for specific words, simple patterns • Good for some things: smok.*, • 25 mg Lasix PO QD • \d+ [um]g [-A-Za-z]+ (PO|IV|IM) (QD|BID|TID|Q6H|Q4H) • dictionary + rules • E.g., names of people, towns, streets, hospitals, clinics, wards, companies; Mr. xxx. • supervised training using single word, bigram, etc., features • mostly leads to probabilistic models that recover the most likely interpretation • parsing to recover syntactic structure of sentences • semantic interpretation in terms of medical vocabularies, taxonomies • discourse analysis for resolution of pronouns, anaphoraExample: Simple text matching • UMLS contains >1M medically meaningful phrases • vocabularies from ~150 sources • e.g.,“heart attack”,“myocardial infarction”,“acute MI”, etc. • synonym, antonym, generalization, specialization, co -occurrence links • 189 semantic types in taxonomy of entities and relations • normalizer, all terms indexed by their normalized versions • Search each of n2 substrings for match in UMLS; then search for best cover by resulting matchesExample: Tawanda Sibanda’s MEng thesis, 2006 http://groups.csail.mit.edu/medg/ftp/tawanda/THESIS.pdf • Tasks: • De-identification: find all of • Patients’ and doctors’ first & last names • Id numbers • Phone, fax, pager numbers • Hospital names • Geographic locations • Dates • Try to resolve ambiguity: • E.g., “Mr. Huntington, who has Huntington’s Disease” • Extract semantic categories • Extract semantic relationsClassifier for De-Id • Features: – Target word to be classified – Words up to 2 words left/right of target – Words up to 2 syntactic links left/right of target (using Link Parser, vide infra) – Target part of speech – Target capitalization – Target length – MeSH ID of noun phrase containing the target – Presences of target ± 1 word in name, location, hospital and monthdictionaries – Heading of document section where target appears – Whether “-” or “/” characters are in target • Support Vector Machine (linear kernel)“Secret Sauce”: Syntax • Link Grammar Parser –Lexical database of constraint formulas for each word (many


View Full Document

MIT HST 950J - Medical Natural Language Processing

Download Medical Natural Language Processing
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Medical Natural Language Processing and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Medical Natural Language Processing 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?