Nico

Nico's Cervical cancer Pathology reports

Notes on data structure with the help of Gemini 2.5 PRO

Data

dt = await (await fetch('https://raw.githubusercontent.com/episphere/gemini/main/doc/Electronic path data exa mple(Sheet1)2_5_pro.json')).json()

Proposed data structure for the 75 reports provided:

{
    "report_id": "...",
    "clinical_history": "...",
    "personnel": [],
    "specimens": [
      {
        "specimen_id": "...",
        "source_procedure": "...",
        "diagnosis": [],
        "gross_description": "...",
        "microscopic_description": "...",
        "comments": []
      }
    ],
    "disclaimer": "...",
    "certification_statement": "..."
  }

short schema metadata

{
  "Chronic cervicitis": {"enum": ["yes","no"]},
  "HPV 18": {"enum": ["yes","no"]},
  "Transformation Zone/Endocervical Glands": {"enum": ["Present","Absent"]}
}

generated schema

{
   "$schema": "http://json-schema.org/draft-07/schema#",
   "title": "Cervical Health Data",
   "description": "A schema for recording basic cervical health indicators.",
   "type": "object",
   "properties": {
      "Chronic cervicitis": {
         "description": "Indicates the presence or absence of chronic cervicitis.",
         "type": "string",
         "enum": [
            "yes",
            "no"
         ]
      },
      "HPV 18": {
         "description": "Indicates the presence or absence of HPV type 18.",
         "type": "string",
         "enum": [
            "yes",
            "no"
         ]
      },
      "Transformation Zone/Endocervical Glands": {
         "description": "Indicates the presence or absence of the transformation zone or endocervical glands.",
         "type": "string",
         "enum": [
            "Present",
            "Absent"
         ]
      }
   },
   "required": [
      "Chronic cervicitis",
      "HPV 18",
      "Transformation Zone/Endocervical Glands"
   ]
}

Wrangling

text to indexed rows

txt = await(await fetch('https://raw.githubusercontent.com/episphere/gemini/main/doc/Electronic%20path%20data%20example(Sheet1).csv')).text()
rows=txt.split(/\r\n/).slice(0,-1) // blank tail removed
rows=rows.map(function(row){return {txt:row,report_id:row.match(/[\w]+/)[0]}})

list possible diagnostics

dt = await (await fetch('https://raw.githubusercontent.com/episphere/gemini/main/doc/Electronic path data exa mple(Sheet1)2_5_pro.json')).json();
diagnostics=[]
specimens=[]
dt.forEach(function(d,i){
    try{
        d.specimens.forEach((sp,j)=>{
            sp.diagnosis.forEach((dia,k)=>{
                diagnostics.push(dia)
            })
        })
    }
    catch(err){
        console.log(err)
    }
})
diagnostics=[...new Set(diagnostics)]

information extraction

const wait = (milliseconds=1000) => {
  return new Promise(resolve => setTimeout(resolve, milliseconds));
};
res=[];
for(var i=0;i<75;i++){
wait();
   res[i] = await (await import('https://episphere.github.io/gemini/extractNico.mjs')).extractNico(i)
}

extract schema from all 75 clinical reports as a single command:

res75 = await (await import('https://episphere.github.io/gemini/extractNico.mjs')).extractNico75()

extracting schema for a particular report (ith):

i=9;
res_i = await (await import('https://episphere.github.io/gemini/extractNico.mjs')).extractNico(i)

checking the session

When running a batch of information extractions by in-browser AI nicoExtract75, you can keep an eye on the shared session. For example, when running extractNico75,

nicoReps = await (await import('https://episphere.github.io/gemini/extractNico.mjs'));

the session is externally available since it is exported (note export { extractNico, extractNico75, session } at the end of extractNico.mjs module). As a consequence, the session is available as nicoReps.session:

session = nicoReps.session

Everytime you call for the session you get something like

LanguageModel {inputUsage: 3081, inputQuota: 9216, topK: 3, temperature: 1, onquotaoverflow: null}

checking the results

The res method was written to record which report (out of 75) was processed. For example, if you want to compare report 9 with report 29,

res = await nicoReps.extractNico(3);
res = await nicoReps.extractNico(8);

you just have to call the res method res

[
   null,
   null,
   null,
   {
      "Chronic cervicitis": "yes",
      "HPV 18": "no",
      "Transformation Zone/Endocervical Glands": "Present"
   },
   null,
   null,
   null,
   null,
   {
      "Chronic cervicitis": "no",
      "HPV 18": "yes",
      "Transformation Zone/Endocervical Glands": "Absent"
   }
]

Show me the data results

If you only came here for the information extraction results then there you have it:

res = await (await fetch('https://raw.githubusercontent.com/episphere/gemini/refs/heads/main/nicoExtracted.json')).json()