Update app.py

app.py

@@ -222,35 +222,44 @@ def log_system_usage(tag=""):
         logger.error(f"[{tag}] Monitor failed: {e}")
 
 def clean_response(text: str) -> str:
-    """Optimized response cleaning with regex compilation"""
+    """Enhanced response cleaning with aggressive artifact removal"""
     if not text:
         return ""
     
-    # Pre-compiled regex patterns
+    # Pre-compiled regex patterns for cleaning
     patterns = [
-        (re.compile(r"\[.*?\]|\bNone\b"), ""),
-        (re.compile(r"To analyze the patient record excerpt.*?medications\."), ""),
-        (re.compile(r"Since the previous attempts.*?\."), ""),
-        (re.compile(r"I need to.*?medications\."), ""),
-        (re.compile(r"Retrieving tools.*?\."), ""),
+        (re.compile(r"\[.*?\]|\bNone\b", re.IGNORECASE), ""),
+        (re.compile(r"To analyze the patient record excerpt.*?medications\.", re.IGNORECASE), ""),
+        (re.compile(r"Since the previous attempts.*?\.", re.IGNORECASE), ""),
+        (re.compile(r"I need to.*?medications\.", re.IGNORECASE), ""),
+        (re.compile(r"Retrieving tools.*?\.", re.IGNORECASE), ""),
+        (re.compile(r"I will start by retrieving.*?\.", re.IGNORECASE), ""),
+        (re.compile(r"This requires reviewing.*?\.", re.IGNORECASE), ""),
         (re.compile(r"\s+"), " "),
-        (re.compile(r"[^\w\s\.\,\(\)\-]"), "")
+        (re.compile(r"[^\w\s\.\,\(\)\-]"), ""),
+        (re.compile(r"(No missed diagnoses identified\.)\s*\1+", re.IGNORECASE), r"\1"),  # Deduplicate
     ]
     
     for pattern, repl in patterns:
         text = pattern.sub(repl, text)
     
-    return text.strip()
+    # Deduplicate identical sentences
+    sentences = text.split(". ")
+    seen = set()
+    unique_sentences = [s for s in sentences if s and not (s in seen or seen.add(s))]
+    text = ". ".join(unique_sentences).strip()
+    
+    return text if text else "No missed diagnoses identified."
 
 def summarize_findings(combined_response: str) -> str:
-    """Optimized findings summarization"""
+    """Enhanced findings summarization for a single, detailed paragraph"""
     if not combined_response:
         return "No missed diagnoses were identified in the provided records."
     
     # Pre-compiled regex patterns
     diagnosis_pattern = re.compile(r"-\s*(.+)$")
     section_pattern = re.compile(r"###\s*(Missed Diagnoses|Medication Conflicts|Incomplete Assessments|Urgent Follow-up)")
-    no_issues_pattern = re.compile(r"No issues identified", re.IGNORECASE)
+    no_issues_pattern = re.compile(r"No issues identified|No missed diagnoses identified", re.IGNORECASE)
     
     diagnoses = []
     current_section = None
@@ -266,7 +275,7 @@ def summarize_findings(combined_response: str) -> str:
             current_section = "diagnoses" if section_match.group(1) == "Missed Diagnoses" else None
             continue
         
-        # Only process diagnosis lines in the correct section
+        # Process diagnosis lines in the correct section
         if current_section == "diagnoses":
             diagnosis_match = diagnosis_pattern.match(line)
             if diagnosis_match and not no_issues_pattern.search(line):
@@ -274,6 +283,25 @@ def summarize_findings(combined_response: str) -> str:
                 if diagnosis:
                     diagnoses.append(diagnosis)
     
+    # Extract findings from non-sectioned text (e.g., psychiatric evaluation, medications)
+    medication_pattern = re.compile(r"medications including ([^\.]+)", re.IGNORECASE)
+    evaluation_pattern = re.compile(r"psychiatric evaluation.*?mention of ([^\.]+)", re.IGNORECASE)
+    
+    for line in combined_response.splitlines():
+        line = line.strip()
+        if not line or no_issues_pattern.search(line):
+            continue
+        
+        med_match = medication_pattern.search(line)
+        if med_match:
+            meds = med_match.group(1).strip()
+            diagnoses.append(f"use of medications ({meds}), which may indicate an undiagnosed psychiatric condition requiring urgent review")
+        
+        eval_match = evaluation_pattern.search(line)
+        if eval_match:
+            details = eval_match.group(1).strip()
+            diagnoses.append(f"psychiatric evaluation noting {details}, suggesting a potential missed psychiatric diagnosis requiring urgent review")
+    
     if not diagnoses:
         return "No missed diagnoses were identified in the provided records."
     
@@ -281,12 +309,11 @@ def summarize_findings(combined_response: str) -> str:
     seen = set()
     unique_diagnoses = [d for d in diagnoses if not (d in seen or seen.add(d))]
     
-    if len(unique_diagnoses) == 1:
-        return f"Missed diagnoses include {unique_diagnoses[0]}"
-    
-    summary = "Missed diagnoses include " + ", ".join(unique_diagnoses[:-1])
-    summary += f", and {unique_diagnoses[-1]}" if len(unique_diagnoses) > 1 else ""
-    summary += ", all of which require urgent clinical review to prevent potential adverse outcomes."
+    # Create a single paragraph
+    summary = "The patient record indicates missed diagnoses including "
+    summary += ", ".join(unique_diagnoses[:-1])
+    summary += f", and {unique_diagnoses[-1]}" if len(unique_diagnoses) > 1 else unique_diagnoses[0]
+    summary += ". These findings, derived from the provided clinical data, suggest potential oversights in the patient's medical evaluation and require urgent clinical review to prevent adverse outcomes."
     
     return summary
 
@@ -312,6 +339,9 @@ def init_agent():
         step_rag_num=4,
         seed=100,
         additional_default_tools=[],
+        disable_tools=True,  # Disable tools to avoid unnecessary calls
+        max_retries=2,  # Limit retries to prevent loops
+        max_tokens=4096,  # Increase token limit for complex inputs
     )
     agent.init_model()
     
@@ -322,7 +352,7 @@ def init_agent():
 def create_ui(agent):
     """Optimized UI creation with pre-compiled templates"""
     PROMPT_TEMPLATE = """
-Analyze the patient record excerpt for missed diagnoses only. Provide a concise, evidence-based summary as a single paragraph without headings or bullet points. Include specific clinical findings (e.g., 'elevated blood pressure (160/95) on page 10'), their potential implications (e.g., 'may indicate untreated hypertension'), and a recommendation for urgent review. Do not include other oversight categories like medication conflicts. If no missed diagnoses are found, state 'No missed diagnoses identified' in a single sentence.
+Analyze the patient record excerpt for missed diagnoses only, focusing on clinical findings such as symptoms, medications, or evaluation results. Provide a concise, evidence-based summary as a single paragraph without headings or bullet points. Include specific clinical findings (e.g., 'elevated blood pressure (160/95) on page 10'), their potential implications (e.g., 'may indicate untreated hypertension'), and a recommendation for urgent review. Do not use external tools unless explicitly required by the excerpt, and avoid mentioning other oversight categories like medication conflicts. If no missed diagnoses are found, state 'No missed diagnoses identified' in a single sentence. Use only the information provided in the excerpt below.
 Patient Record Excerpt (Chunk {0} of {1}):
 {chunk}
 """
@@ -352,7 +382,6 @@ Patient Record Excerpt (Chunk {0} of {1}):
             file_hash_value = ""
             
             if files:
-                # Use cached results when possible
                 for f in files:
                     file_type = f.name.split(".")[-1].lower()
                     cache_key = f"{file_hash(f.name)}_{file_type}"
@@ -426,7 +455,7 @@ Patient Record Excerpt (Chunk {0} of {1}):
                                         elif isinstance(chunk_output, str):
                                             content = clean_response(chunk_output)
                                         
-                                        if content:
+                                        if content and content != "No missed diagnoses identified.":
                                             chunk_response += content + " "
                                 
                                 if chunk_response:
@@ -434,7 +463,6 @@ Patient Record Excerpt (Chunk {0} of {1}):
                                     history[-1] = {"role": "assistant", "content": combined_response.strip()}
                                     yield history, None, ""
                             finally:
-                                # Ensure cleanup
                                 del future
                                 torch.cuda.empty_cache()
                                 gc.collect()
@@ -458,7 +486,6 @@ Patient Record Excerpt (Chunk {0} of {1}):
                 history.append({"role": "assistant", "content": f"❌ Error occurred: {str(e)}"})
                 yield history, None, f"Error occurred during analysis: {str(e)}"
             finally:
-                # Final cleanup
                 torch.cuda.empty_cache()
                 gc.collect()